Most teams publish a policy once and then stop looking at it. A phone number in the directory changes, a new sending service goes live, and the record at the DNS host still describes last year’s setup.

A DMARC record is the DNS TXT value at _dmarc. It names the policy for mail that fails alignment, and the mailboxes that should receive reports. The generator on this page loads the currently published record, or accepts a pasted value, and rebuilds the line as you change it. Copy the result into DNS yourself. The page does not edit your domain.

Click for DMARC Record Generator

Change the Policy Before You Publish It

Start from the record the lookup returns. If the policy is p=none, receivers are being asked to deliver the mail and send reports. Quarantine asks them to treat a failure as spam. Reject asks them to refuse it. Move from none to a stricter policy only after the reports show that your own mail is aligned.

Alignment is the part people skip. A message passes when SPF or DKIM passes, and that passing domain matches the visible From domain. Relaxed alignment, the default, allows a shared organizational domain. Strict alignment requires an exact match. The percentage tag applies the policy to a sample of failing mail, letting you test quarantine or reject before it affects every failure.

Aggregate reports, called RUA, are the daily XML files. Failure reports, RUF, are the per-message notes, and many receivers send few of them. A logo in the inbox depends on BIMI, which can’t display when the authentication under it is missing.

The Other DMARC Guides

The generator is the record itself. These pages cover the reports, the neighboring records, and the setup checks around it.

• What is DMARC? An overview of DMARC and why it’s necessary for email authentication.
• DMARC RUA Report Viewer reads an aggregate report and checks whether receivers followed the policy in that file.
• Import, Modify, or Build Your SPF Record from Scratch is the matching builder for the SPF record alignment that depends on.
• What is Email Authentication? SPF, DKIM, and DMARC Explained puts the three checks in one place.

Look up the record, change only the tags you mean to change, and publish one TXT value at _dmarc. Then read the next report before you tighten the policy again.

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: App: Import, Modify, or Build Your DMARC TXT Record from Scratch

Every employee email leaves with a signature, and most of those signatures were pasted in by hand. Phone numbers go stale. Legal lines differ by office. A campaign banner shows up in one team’s mail and nowhere else.

Marketing wants a banner or a survey in the signature this month. The technology group wants one approved layout. Neither side should have to open every mailbox to get there.

Exclaimer is a cloud platform that designs an email signature once and applies it to Microsoft 365 and Google Workspace mailboxes. IT publishes the template. Employees do not rebuild it on their own computers.

Trusted by 80,000+ businesses and 9 million email accounts worldwide. Exclaimer created the first-ever email signature software in 2001. We’re in over nine million outboxes, at companies like Sony, Mattel, Bank of America, NBC, the Government of Canada, and the BBC.

Exclaimer, company homepage

A single template keeps the logo, the colors, and the contact fields in one place. When a phone number changes in the directory, the signature can follow that record instead of waiting on a manual paste.

Marketing can schedule a banner, rotate a campaign, or add a survey without sending a new image to every employee. The signature stays a company asset, not a personal footer.

What used to take weeks now gets done in 15 minutes. Our team can easily keep branding consistent and every email looks polished. Plus, marketing now runs dynamic campaigns directly through our email signatures.

Jessica C. Brandt, VP of Customer Marketing, Bloomerang

What One Template Can Carry

The designer is a drag-and-drop layout, so a team does not have to write HTML for a standard signature. The controls below are what that template can hold once the layout exists.

• Accessibility Controls: Accessibility Controls add a screen-reader (ARIA) label to a link, declare the signature’s language, and keep contact details as real text with alt text on images. Exclaimer built the controls for WCAG 2.1 Level AA checks that apply to a signature. The company’s product FAQ says the tool does not, by itself, make an organization compliant. Public-sector web rules in the United States are moving toward that same level, and those rules name websites and apps rather than a deadline for the signature. A QR code in the signature can also be clicked, so a recipient who does not scan it still reaches the same place.
• Analytics: On the Pro plan, engagement dashboards and one-click surveys show whether a banner or a survey was used.
• Brand Kits: Approved logos, colors, and fonts are stored once and reused on every signature. Exclaimer’s accessibility guide asks teams to check those color pairs for contrast before saving a kit.
• Campaign banners: Standard and Pro plans can schedule, rotate, and target a banner. Pro can also place a social feed, including a LinkedIn feed, in the signature.
• Directory sync: Name, title, and phone can come from the company directory, with a controlled editor for fields such as pronouns.
• Integrations: The Pro plan includes a Salesforce connection. It also supports HubSpot and Calendly.
• Legal disclaimers: Disclaimer text is managed apart from the signature layout on Standard and Pro, so a legal line can change without redrawing the card. That option is not part of the client-side Google Workspace setup.
• Meeting Branding: The Pro plan applies video backgrounds and name tags for Teams, Zoom, and Google Meet from the same brand kit.
• QR codes: A code can point to a vCard, a booking page, or a campaign link. A vCard is included on Pro.
• Role-based rules: Apply a template by user, department, role, location, or recipient. Marketing can edit within limits that IT sets.
• Template designer: Drag-and-drop blocks and a template library. Starter includes one template, Standard adds a higher cap, and Pro removes the fixed cap.

Together, the template holds the signature, disclaimer, campaign, and brand kit. IT publishes it once instead of passing an image file from mailbox to mailbox.

Accessibility programs almost always start with the website. Email signatures get missed because nobody thinks of them as digital content, but they are, and they appear in thousands of everyday business communications. With Accessibility Controls, IT can set the standard once, at the template level, instead of asking every employee to fix their own signature.

Paul Hammond, Chief Product and Technology Officer, Exclaimer

How a Rollout Works

Teams connect Microsoft 365 or Google Workspace, sync the directory, and choose server-side imprinting or a client-side add-in. Server-side applies the signature for every mail client, including mobile. The Outlook add-in covers client-side use on Microsoft 365. A published template, including its accessibility settings, applies to the group without visiting each mailbox.

Plans are priced per active mailbox, with a minimum of 10 users. A 14-day trial covers the Standard plan. Starter, Standard, and Pro add templates, campaigns, and analytics.

Request an Exclaimer Demo

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: Exclaimer: One Template Applied For Every Email Signature in Your Organization

As of August 24, 2026, public pricing across major MaaS platforms—including DeepSeek, OpenRouter, DeepInfra, and Together AI—combined with model benchmark data from Artificial Analysis, shows that Bitdeer AI Model Studio is the lowest-priced provider for the DeepSeek V4 Flash 0731 API.

On Bitdeer AI Model Studio, DeepSeek V4 Flash is priced at $0.042 per million input tokens, $0.009 per million cached input tokens, and $0.084 per million output tokens.

With an MoE architecture and long-context support, DeepSeek V4 Flash suits coding agents, RAG, and multi-step agent workflows. These applications repeatedly call system prompts, tool definitions, and shared context, so input, cached-input, and output costs directly affect API spending.

How Large Is the DeepSeek V4 Flash API Price Gap?

The comparison combines input, cached-input, and output pricing, using Artificial Analysis’ 7:2:1 cached-input/input/output ratio to calculate blended calling costs.

Under the 7:2:1 ratio, Bitdeer AI’s blended cost is about $0.023 per million tokens, below the blended costs of OpenRouter’s current lowest standard route, DeepInfra, Together AI, and DeepSeek’s official off-peak pricing.

DeepSeek V4 Flash API costs are therefore better assessed through the blended cost of a complete workload rather than any single token rate.

Developers looking to benchmark or test the API could take advantage of limited-time trial offers. Bitdeer AI previously offered new users $5 in Model Studio Credit for Featured Models who registered from August 17 to September 15, 2026, with the credit valid for 90 days from issuance. Model Studio has also periodically offered discounts on popular models.

Real-World Scenario: How Much Can a Production Agent Save Each Month?

Assume a coding agent or enterprise RAG system consumes 1 billion tokens per month at the same 7:2:1 ratio:

• 700 million tokens come from cacheable system prompts, tool definitions, codebases, or knowledge-base context;
• 200 million tokens are new inputs;
• 100 million tokens are model outputs.

Compared to Together AI, Bitdeer AI saves developers about $53.9 per billion tokens. At a scale of 10 billion tokens per month, that margin expands to about $539.

For high-volume workloads such as multi-agent orchestration, long-context coding, batch content production, or data processing, this difference directly affects operating costs.

Lower unit costs also expand the agent iteration budget, allowing more tool calls, retries, reflection rounds, and retained context within the same budget.

Does Lower Pricing Mean Slower Speed?

Does a price roughly 70% lower than others imply compromised latency or throughput?

The Artificial Analysis benchmark snapshot from August 24, 2026, shows Bitdeer AI in its Most Attractive Quadrant, with low blended token pricing and competitive end-to-end response time.

Bitdeer AI delivers ultra-low token pricing without sacrificing inference throughput or response time. For continuously running coding agents, RAG, and multi-step workflows, this price-performance balance may matter more than chasing the lowest price or fastest response alone.

Don’t treat public benchmarks as an SLA. Actual inference performance can vary with request length, concurrency, testing region, and platform load. Teams moving production traffic should stress-test their prompt lengths, concurrency levels, user regions, and time-to-first-token requirements before selecting a provider.

Beyond Price, Why Does Infrastructure Matter?

As call volumes grow, another question matters: does a platform have sufficient compute and resource-scheduling capacity to absorb the traffic?

Bitdeer AI Model Studio does not resell third-party APIs; it runs on Bitdeer’s own data centers and NVIDIA GPU infrastructure.

As of July 2026, Bitdeer AI had deployed 4,248 H100, H200, B200, GB200, and GB300 GPUs, with AI Cloud utilization reaching 95%. Integrated infrastructure spanning compute, resource scheduling, and the Model Studio API can help improve GPU utilization, optimize inference costs, and support sustained throughput and stability.

For production environments, lower pricing creates lasting value only with sufficient compute capacity and reliable delivery.

From Model Coverage to Inference Economics and Stability

Competition in the open-source model API market is shifting from how many models are supported to whether providers can deliver model capabilities at lower cost and with greater stability.

Bitdeer AI Model Studio is the MaaS layer within Bitdeer’s full-stack AI Cloud, supported by its own NVIDIA GPU infrastructure and ongoing inference optimization. For developers, Model Studio is more than an API endpoint; it offers an integrated service capability from compute to model delivery.

In practice, this infrastructure supports the MaaS service by helping control unit token costs while providing capacity, availability, and scalability as workloads expand.

For production applications such as coding agents, RAG, and multi-agent workflows, the key question is no longer simply whether a platform offers DeepSeek, but whether the same model can run continuously at a more reasonable cost and with more stable performance.

As more platforms offer the same open-source models, differentiation will increasingly depend not on model coverage alone, but on the ability to deliver model capabilities economically, reliably, and over the long term.

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: Which DeepSeek V4 Flash API Is Cheapest? Bitdeer AI Tops Artificial Analysis Price Benchmark

People usually write the argument for a pitch, lesson, or quarterly review before anyone opens a slide file. It sits in meeting notes, a lesson plan, or a hastily drafted document. The slow part is breaking that material into slides a room can follow. Layout and consistency take the hour that should have gone to the point.

A second delay shows up when the same story has to exist in more than one program. One person presents from a file on a laptop. A teammate expects a shared deck online. Rebuilding both by hand turns a clear idea into a late file.

SlidesAI

SlidesAI is an AI presentation platform that builds an editable deck from a prompt, pasted notes, a document, or a web page. You can start in the browser, inside Google Slides, inside PowerPoint, or from a ChatGPT conversation that saves a Google Slides file.

Keep Editing Where the Deck Will Be Given

The useful result is not a stack of slides you cannot change. You review an outline first. The deck is then generated in the file you’ll keep editing. In the browser, you export the file when it is ready. In Google Slides or PowerPoint, new slides land in the file you already have open and follow that file’s theme.

A class, a sales review, and a client pitch do not have to start in the same program. Notes can begin in the browser. A deck that already lives in PowerPoint can take new slides on top of its Slide Master. A conversation in ChatGPT can end as a Google Slides file.

Spend 85% less time building slides. 3.7M+ presentations created with SlidesAI, and counting.

SlidesAI

What You Can Change Before You Present

The work is built around the deck, not around a single add-on menu. These are the controls you use while a presentation is still taking shape.

• AI image generation: Describe a picture or icon, then place the result on a slide.
• Brand kit: Save colors, fonts, a logo, and layouts, including a look taken from a PowerPoint file you upload.
• Charts and diagrams: Build charts from your own figures, and add diagrams, flowcharts, and timelines you can still edit.
• Document import: Bring in a Microsoft Word document, a PDF, a Google Doc, or a page from the web. Document upload is included on the paid plans.
• Editable outline: Read the outline and change it before any slides are built.
• Export: Send the deck to Google Slides, download a PowerPoint file, or save a PDF.
• Live collaboration: Co-edit, comment, and share a link from the browser app. A Google Slides deck keeps Google’s own sharing.
• Multilingual decks: Generate a presentation in another language, or translate a deck you already have.
• Remix and rewrite: Shorten, expand, or rephrase slides that are already in the file.
• Speaker notes: Ask the PowerPoint add-in to draft notes under the slides.
• Stock images: Search royalty-free photos and drop them in, or upload your own.
• Templates: Start from a library of Google Slides and PowerPoint templates when you do not want a blank file.
• Theme controls: Restyle an existing Google Slides or PowerPoint deck in one pass, and keep custom themes for the next one.

Taken together, the first draft stays editable. You are not locked to the first layout, the first wording, or the first program that opened the file.

How to Create a Presentation

Pick the place the deck needs to end up. Each step is short.

1. In the browser, open SlidesAI with no install. Enter a topic or paste notes, or upload a document. Set the type of presentation, the audience, the tone, and the style. Review the outline, generate the slides, then edit them and export to Google Slides, PowerPoint, or a PDF.
2. In Google Slides, install the SlidesAI add-on, open a deck, and generate slides into that file. The new slides follow the deck’s theme and layouts. Later edits stay in Google Slides. A personal Google account and a Google Workspace account both work.
3. In PowerPoint on Windows or Mac, install the SlidesAI add-in and generate into the open deck, including onto your Slide Master. PowerPoint 2016 and newer are supported, as are the Microsoft 365 desktop apps. Generation can add speaker notes.
4. In ChatGPT, open the SlidesAI presentation maker, describe the deck, and review the outline. Sign in with Google. The file is saved to Google Slides.

Each path ends in a deck you can still change. The browser is where you start from scratch. Google Slides and PowerPoint are where you can generate a file that already has a look. ChatGPT is where you describe the presentation before it becomes a Google Slides file.

Create a Presentation with SlidesAI

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: SlidesAI: Build Decks in the Browser, Slides, or PowerPoint

Before investing in dashboards, machine learning, or advanced attribution models, marketing teams should ask a simpler question: Do we truly understand our data? Exploratory Data Analysis (EDA) provides that understanding by transforming raw marketing data into actionable business insights.

Marketing Generates Data. EDA Generates Understanding

Modern marketing teams collect data from almost every customer interaction.

Website analytics reveal how visitors navigate digital properties. Advertising platforms measure impressions, clicks, conversions, and return on ad spend. CRM systems track opportunities through the sales pipeline. Marketing automation platforms record lead generation and nurturing activities.

The challenge is no longer collecting information – it’s making sense of it.

Many organizations invest heavily in dashboards that display dozens of metrics but still struggle to answer fundamental business questions:

• Which campaigns are actually driving revenue?
• Why are conversion rates declining?
• Which marketing channels deserve more investment?
• Are qualified leads coming from the right acquisition sources?

These questions require more than visualization. They require investigation.

That’s where Exploratory Data Analysis (EDA) becomes one of the most valuable practices in Marketing Analytics.

What Is Exploratory Data Analysis?

Exploratory Data Analysis is the process of examining datasets before applying predictive models or building executive dashboards.

Rather than immediately creating reports, analysts first seek to understand:

• Data quality
• Missing values
• Variable distributions
• Relationships between metrics
• Trends over time
• Unexpected patterns
• Potential anomalies

The goal is simple:

Transform raw data into reliable business knowledge.

Without this step, organizations risk making decisions based on incomplete, inconsistent, or misunderstood information.

Why Marketing Teams Should Care About EDA

Marketing data rarely comes from a single source.

A typical organization may use:

• Google Analytics 4
• Google Ads
• Meta Ads
• HubSpot CRM
• RD Station
• Email Marketing platforms
• E-commerce systems

Each platform measures performance differently.

An Exploratory Data Analysis helps connect these sources while identifying inconsistencies before they affect business decisions.

For example, an EDA might reveal:

• Campaigns with high click-through rates but low conversion value
• Organic traffic generating more qualified leads than paid campaigns
• Lead generation assets outperforming direct sales offers
• CRM stages accumulating opportunities without progression
• Marketing campaigns missing attribution parameters

These insights often remain hidden inside dashboards focused only on performance metrics.

From Metrics to Business Questions

One of analytics’ biggest misconceptions is that the work begins with SQL queries or visualization tools.

In reality, effective analyses begin with business questions.

EDA encourages analysts to organize investigations around decision-making rather than reporting.

That shift changes the value marketing teams receive from analytics.

Best Practices for Marketing EDA

Organizations can significantly improve analysis quality by following a structured exploratory process.

1. Start with Data Quality

Before calculating KPIs, validate the dataset.

Check for:

• Missing values
• Duplicate records
• Incorrect data types
• Inconsistent campaign names
• Null attribution fields

Good decisions require trustworthy data.

2. Understand the Business Context

Numbers rarely explain themselves.

A 5% conversion rate may be excellent for one campaign and disappointing for another.

Every analysis should begin with understanding:

• Marketing objectives
• Customer journey
• Acquisition channels
• Campaign strategy

Context transforms metrics into insights.

3. Analyze Before Building Dashboards

Dashboards summarize information.

EDA discovers information.

Before creating visualizations, explore:

• Trends
• Distributions
• Correlations
• Outliers
• Segment performance

Dashboards become much more meaningful when built on exploratory findings.

4. Focus on Decisions, Not Reports

Every analysis should answer a practical question.

Examples include:

• Should we increase investment in Google Ads?
• Which lead magnets deserve additional promotion?
• Are remarketing campaigns generating profitable customers?
• Which sales pipeline stages need attention?

If an analysis doesn’t influence a decision, its business value is limited.

A Practical Marketing Analytics Workflow

An effective exploratory workflow often follows this this sequence:

1. Define the business problem.
2. Assess data quality.
3. Explore variables and distributions.
4. Identify relationships between marketing metrics.
5. Generate business insights.
6. Recommend actions.
7. Build dashboards only after understanding the data.

This approach reduces the risk of presenting misleading metrics while increasing confidence in strategic recommendations.

A Real-World Use Case

While developing an end-to-end Marketing Analytics portfolio, I explored datasets covering multiple stages of the customer journey, including:

• Website Analytics
• Facebook Ads
• Google Ads
• CRM Sales Pipeline
• Marketing Automation

Rather than limiting the work to descriptive dashboards, I investigated each dataset through Exploratory Data Analysis.

This process revealed patterns such as:

• Educational content generating significantly more lead conversions than direct commercial offers.
• Google serving as the strongest acquisition source across multiple marketing initiatives.
• Lead-nurturing campaigns consistently outperformed isolated promotional campaigns.
• Sales opportunities concentrating in specific CRM stages.
• Different advertising platforms exhibiting distinct relationships between spend, conversion value, and customer acquisition.

These findings were only possible because the analysis prioritized business questions before visualization.

EDA as a Competitive Advantage

Marketing technology continues to evolve rapidly.

Artificial Intelligence can automate reports.

Dashboards update in real time.

Attribution models become increasingly sophisticated.

Yet none of these capabilities replace critical thinking.

Organizations that consistently make better marketing decisions usually excel at asking better questions before analyzing data.

Exploratory Data Analysis strengthens that capability.

It enables teams to move beyond describing performance to understanding why it changes, and what should happen next.

Conclusion

Marketing teams don’t need more dashboards.

They need greater confidence in the decisions those dashboards support.

Exploratory Data Analysis provides that confidence by connecting technical analysis with business strategy.

Before optimizing campaigns, launching attribution models, or implementing predictive analytics, organizations should invest time in understanding their data.

Because better marketing decisions begin long before the final report is created.

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: How Exploratory Data Analysis Strengthens Marketing Decision-Making

$13 billion. That’s how much advertisers wasted on Made-for-Advertising sites back in 2023 alone, according to the Association of National Advertisers. Not fraud in the usual sense. Not malware or bot farms. Just pointless websites designed to bleed programmatic ad budgets.

If you’re buying programmatic ads, some of your budget is almost certainly going to these sites right now.

What MFA Sites Are

MFA sites aren’t what most people think when they hear ad fraud. They’re not serving malware. They’re not spoofing domains. They don’t even technically violate most platforms’ policies.

They’re just completely worthless.

An MFA site exists for one reason: to siphon money from programmatic ad budgets. The model is simple. Publish low-effort content — often scraped or machine-written — pack the pages with ads, buy cheap traffic through arbitrage, and pocket the spread between traffic costs and ad revenue.

MFA sites comprised 21% of all programmatic ad impressions and attracted 15% of ad spend. The ANA estimated this translates to roughly $13 billion in wasted ad spend annually.

Association of National Advertisers

That’s not a typo. Thirteen billion dollars went to sites that provide zero value to anyone except their operators.

How They Work

The typical MFA operation looks like this:

They create hundreds or thousands of websites with thin content. 10 Things You Didn’t Know About Coffee or Is It Going to Rain Tomorrow? The content isn’t completely fake—it’s just useless, often generated in bulk.

They buy cheap traffic. Google Ads, Facebook Ads, native ad networks, anywhere they can acquire clicks for $0.01-0.05. Sometimes they use misleading headlines or thumbnails. Sometimes they just rely on accidental clicks.

They load pages with ads. Lots of ads. The ANA’s definition of MFA sites notes they typically exhibit ad-to-content ratios of 30% or higher for desktop—usually at least twice the internet average. Auto-refresh ads, infinite scroll, aggressive ad placement—anything to maximize impressions.

They collect programmatic ad revenue. Because they’re buying traffic at $0.03 and selling ad impressions at $0.10, they profit. Scale this across thousands of sites and millions of impressions, and you’ve got a business model.

None of this is obviously fraudulent. The sites are real. The traffic is technically human (even if it’s accidental). The ads are served. Brand safety tools don’t flag them because there’s no objectionable content. Viewability metrics look fine because the ads do load on real browsers.

The Scale of the Problem

Only 36 cents of every dollar entering a demand-side platform effectively reaches the end consumer. The rest is lost to a “cost waterfall”—29% goes to ad-tech transaction costs, while 35% is wasted on unmeasurable or low-value environments like MFA sites.

ANA’s December 2023 Report

For a brand spending $10 million on programmatic, a 15% MFA rate means about $1.5 million ends up on sites that don’t drive any real business outcome. In some cases, the share is much higher — the study found advertisers sending up to 42% of their budgets to MFA domains.

MFA sites have also become harder to spot. Early ones were easy to spot: thin content, pages overloaded with ads, and questionable domains. Newer operations are more careful. They buy expired domains with existing authority, publish more content to look credible, and keep ad density low enough to stay under automated detection.

A 19% year-over-year increase in MFA impression volume in 2023, primarily driven by generative AI making it easier to create content at scale.

DoubleVerify’s 2024 Global Insights Report

Why Your Current Protections Aren’t Working

Brand safety tools are designed to protect you from appearing next to objectionable content—hate speech, adult content, violence. MFA sites don’t have objectionable content. They have boring, inoffensive, useless content. Your brand safety tool sees “10 Things You Didn’t Know About Coffee” and gives it a pass.

As the ANA noted in its MFA definition report, MFA websites show high measurability and viewability, low invalid traffic, and usually brand-safe environments. They’re designed to fool digital advertising buyers.

Inclusion lists help, but most advertisers don’t use them aggressively enough. The ANA found that only 46% of marketers even track how many websites they use for programmatic campaigns. The average campaign runs on 44,000 websites—far more than necessary to reach quality audiences.

Viewability metrics don’t help because MFA sites actually have decent viewability. The ads load. They’re technically visible. The fact that nobody’s actually paying attention doesn’t show up in the data.

Fraud detection tools miss them because there’s no fraud in the traditional sense. Real browsers, real page loads, real impressions. Just zero value.

The Economics That Keep MFA Alive

MFA sites exist because the economics of programmatic advertising make them profitable.

Traffic arbitrage still works. You can still buy clicks for less than you earn from serving ads to those clicks. As long as that equation holds, MFA sites will exist.

Programmatic advertising prioritizes scale and efficiency over quality. DSPs optimize for reach and cost-per-impression. They’re not designed to distinguish between an impression on The New York Times and an impression on “toptennewstoday.info” if both cost the same and meet basic quality thresholds.

MFA sites captured 11% of open web programmatic ad dollars in Q3 2024, representing a multi-billion-dollar problem that continues to grow despite industry efforts.

eMarketer

What You Can Do

Keep exclusion lists tight and up to date. If a site exists mainly to run ads, block it. Companies like DoubleVerify and Integral Ad Science now offer MFA-specific detection with tiered classification systems, but you need to actively use these tools and update your lists regularly.

Go through your programmatic buys every month. Check placement reports and confirm where ads actually appeared. Review the domains. Visit the sites yourself. If you’re seeing low-effort content packed with ads, block those placements. It’s repetitive work, but it’s the only reliable way to track where your budget ends up.

Put a larger share of spend into PMPs and direct deals with publishers you know. MFA activity is heavily concentrated in the open exchange. Private deals usually come at a higher CPM, but they give you access to real audiences instead of resold traffic.

Demand transparency from your DSP. If your demand-side platform can’t or won’t tell you exactly where your ads ran, that’s a problem. The ANA found that data access is lacking across the industry, resulting in transparency issues that lead to inefficiencies and waste.

Use attention metrics, not just viewability. Viewability tells you an ad loaded. Attention metrics tell you if anyone actually looked at it. MFA sites have terrible attention metrics because even though ads are viewable, nobody’s engaging with the content or ads.

What Publishers and Platforms Need to Do

If you’re a publisher or ad tech platform, MFA sites are your problem too. They’re devaluing programmatic inventory across the board.

At Axis, we see this from the supply side. Publishers ask us why their CPMs are falling, and part of the answer is that MFA sites are flooding the market with cheap, worthless inventory that drags down pricing for everyone.

The industry needs better standards for what qualifies as legitimate publisher inventory. The IAB Tech Lab has worked on transparency initiatives, but we need more aggressive filtering at the exchange level.

SSPs need to be more willing to cut off MFA inventory even when it’s technically compliant. Yes, it means less volume. But protecting programmatic advertising’s value is worth the short-term revenue hit.

Publishers need to advocate loudly for quality standards that exclude MFA. Every dollar that goes to an MFA site is a dollar that didn’t go to a legitimate publisher creating actual content.

The Numbers Don’t Lie

MFA sites are a symptom of programmatic advertising that optimized for efficiency and scale without adequately protecting for quality. We built systems that could buy and sell billions of impressions per day, but we didn’t build adequate safeguards to ensure those impressions had any value.

MFA sites pulled in about $13 billion that advertisers expected to reach legitimate publishers and real audiences. Instead, a large share of that budget went to pages built mainly to serve ads, with little chance of driving meaningful results.

Cutting this down requires stricter rules on where ads can appear, stronger quality filters across platforms, and a clear understanding that not all impressions carry the same value, even when they pass the same technical checks.

If you buy programmatic media, assume some of your budget is landing on MFA sites right now. Pull the reports. Check the domains. Build and maintain exclusion lists. It’s dull work, but it’s the only way to stop paying sites that add no value.

MFA won’t disappear on its own. The incentives favor the operators, and the system makes it easy. With consistent effort, though, you can cut down how much of your budget vanishes into it.

Your CFO will thank you. So will the legitimate publishers competing for the same ad dollars.

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: Made-for-Advertising Sites Are Stealing Your Money

I work across three screens, and I take a ton of screenshots. A campaign review, a broken layout, a slide someone asked for in chat: each one lands on the Desktop, and by the afternoon I cannot tell them apart.

Taking the shot is the easy part. Finding the one I need, then getting rid of the rest, is what I dread. The cleanup is always the same. I open the folder, sort by name, select from the first screenshot to the last, right-click, and move the whole range to the Trash.

The system already knows how to take the picture. I wanted a place for those pictures to wait, large enough to recognize, close enough to copy into an email or ticket, and easy to clear when the work is done.

I built ShotPanel, a free native app for macOS 14 and later. It gathers the screenshot files macOS already saves into one floating panel, so those images do not cover my Desktop.

The Idea Behind the Panel

Tendedero, an open-source macOS app by Alejandro Buján, was the spark. Tendedero’s clothesline was the idea I wanted: screenshots should gather in one place instead of covering the Desktop. ShotPanel is a totally separate app, but his work inspired it.

I wrote ShotPanel entirely with Grok. I pointed Grok at Tendedero’s GitHub repository, then ran dozens of tests and optimizations with Grok myself. Grok also produced the overview video.

The film below is the 48-second overview that ships with version 1.1.0. Grok produced it.

A Calmer Place for the Shots I Take

Finding the shot was the slow part, not taking it. I was tired of scrolling through a Desktop full of exports, logos, and downloads that happen to be images. The panel holds only the automatic screenshots.

ShotPanel keeps new shots from landing on my Desktop. Files already in the screenshot folder move into the panel, and the floating thumbnail steps aside while that option is on. When I turn the option off or quit, the old save location returns.

From the panel I can open a shot, copy it, or drag that one file into a browser, a ticket, or a folder. Clearing one image, or the whole group, asks first and then moves the files to the Trash. Nothing is deleted outright.

The same group also appears as a ShotPanel folder in the Dock, just after Downloads, for as long as a screenshot is in the panel. I can reach the files without opening a folder and sorting by name, even when the panel itself is minimized.

What the Panel Can Do

I kept the controls small on purpose. Each one does a single job with the files already in the screenshot folder.

• Arrow keys: Pointing at the panel shows arrows on the ends that still have shots. The arrow keys move the row.
• Copy: Puts the selected image on the clipboard. The button then says Copied.
• Delete: Asks Are you sure? on that image, then moves that one file to the Trash.
• Delete all: Asks Are you sure? in the bar, then moves the whole group to the Trash. The same action is on the menu-bar icon.
• Desktop capture: While ShotPanel is open and this option is on, shots already in the screenshot folder move into the panel, new ones skip the Desktop, and the floating thumbnail is turned off. Quitting, or turning the option off, restores the previous save location.
• Dock folder: While any screenshot is in the group, a ShotPanel folder sits in the Dock just after Downloads.
• Drag out: A drag that starts on a photo moves that one file into another app, a browser, or a folder.
• Full-screen hide: The panel hides while a Space is in full screen.
• Horizontal or vertical: Right-click to stack shots upright or sideways. Vertical swaps the panel’s width and height. Each orientation remembers its own size. The info bar stays at the bottom.
• Info bar: Shows the selected shot’s name, date and time, pixel size, and file size.
• Menu bar: Hide or show the panel, turn Desktop capture on or off, open at login, delete the group, or quit.
• Minimize: The minus button, or Command-M, sends the panel to the Dock.
• Move: Drag the bottom bar to move the panel.
• Newest first: The newest shot sits at the start of the list, large enough to read.
• Open: Double-click, or right-click and choose Open, to open the file.
• Open at login: ShotPanel asks before the panel opens. The answer can be changed later from the ShotPanel menu.
• Quit: From the menu bar item, with Command-Q, or by right-clicking the panel.
• Resize: Drag an edge or a corner. The panel keeps that size, and the shots scale to it. One shot scales down so the whole image stays inside. When several fit, they sit side by side with space between them, and the next shows as a faded sliver at the edge.
• Screenshot folder permission: The app asks to read the screenshot folder and stays closed until that is allowed.
• Select: Click a shot. Copy and Delete appear on that image.
• Slide: A finger on the top of the mouse, or a press-and-slide on the trackpad, moves through the list. Shots enter and leave at the edges.

Taken together, the panel is the holding shelf I wanted. I open a shot, copy it, drag it into the app that needs it, or send the group to the Trash, without sorting a folder by name and selecting from the first shot to the last.

Install It and Take a Screenshot

Download the disk image, open it, and drag ShotPanel to Applications. The first launch may say macOS cannot verify the developer. Allow from System Settings:

• Click Open Anyway and enter your Mac’s administrator password to confirm.
• Try to open the app normally. It will be blocked with a warning message.
• Open System Settings from the Apple menu.
• Click on Privacy & Security in the sidebar.
• Scroll down to the Security section. You will see a message that the app was blocked.
• Click Allow.

Right-click ShotPanel, choose Open, then Open again. The app asks to read the screenshot folder and whether to start at login. During screen sharing, that question can appear on the computer itself.

I still take a screenshot the way I always have. With Desktop capture on, the new file shows up in the panel instead of on the Desktop. I double-click it to open it, or select it and copy it.

The source is Swift on macOS 14 or later. A build script in the repository produces the app. A local build is signed just enough to run, and macOS may ask the same questions before the app can read the Desktop.

Download ShotPanel

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: ShotPanel: Organize and Keep Screenshots Off Your Mac Desktop

Moving a sentence from one system to another is rarely a clean copy and paste. A header, a query string, a data URI, or a configuration value accepts a narrow set of characters. A line break, a quotation mark, or an emoji is enough to split the value or get it rejected. The practical repair is to rewrite the text as a string made only of letters, numbers, and a few symbols, then turn that string back into the original words when you need them.

Encode and Decode in the Browser

Base64 is that encoding. Paste ordinary text and this page turns it into a Base64 string. Paste a Base64 string and it turns the string back into text. Both steps run in the browser, using UTF-8, so accents and emoji survive the round trip.

Click for Base64 Encode / Decode

Why a Local Encoder Helps

You can check a value before it goes into a template, a pixel, or a header. If a string you were given looks like noise, decode it on the spot and read what was stored. Nothing is uploaded. The source stays in the page, which matters when the text is a credential, a customer note, or a draft you do not want in someone else’s converter.

Decoding also forgives the mess that real strings pick up. Line breaks inserted by mail clients are ignored. The address-safe alphabet is accepted, where plus and slash were swapped for a hyphen and an underscore, and missing padding is restored before the text is read back.

What the Encoder Does

The controls cover the whole round trip, from the first paste to the copied result.

• Clear failure message: a string that is not valid Base64, or text the browser cannot read, is reported in the result instead of a silent blank.
• Copy in one click: the Copy button, or a click in the result field, puts the output on the clipboard.
• Decode a wrapped string: whitespace and line breaks are stripped, the address-safe alphabet is converted back, and missing padding is added.
• Encode Unicode text: characters outside basic Latin, including emoji, are encoded as UTF-8 bytes before they become Base64.
• Stay in the browser: encode and decode run on the page. The source is not posted to a server.

Taken together, the page lets you prove an encoding, recover a string you were sent, and copy a clean result without leaving the article.

How to Use It

Paste the text you want to send, or the Base64 string you want to read, into Source Data. Choose Base64 Encode or Base64 Decode. The output lands in Result Data, ready to copy.

Keep the page open while you build the header, the data URI, or the configuration value. Encode, place the string where it has to travel, and decode it again if you need to confirm nothing was lost.

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: Base64 Encode and Decode: Make Any Text Safe to Paste

Marketing teams spend weeks planning a webinar and then treat it as a one-hour event. The slides go into a folder, the recording sits in a cloud drive, and the registrant list gets one follow-up email. Most of the value is left on the table.

The better way to think about a webinar is as a content engine. One well-run session can feed your email program, your social calendar, your blog and your sales team for a month. The tool you run it on decides how much manual work that takes.

In this article, we look at that workflow end to end, using Restream’s webinar platform as the example. Restream is best known for multistreaming, sending one live stream to YouTube, LinkedIn, Facebook and other channels at once. Its webinar product, Restream Webinars, adds registration, a live studio in the browser, recording, and analytics on top.

Step 1: One Link for the Whole Lifecycle

The first friction point in any webinar is the attendee journey. Restream uses a single link for registration, the live session and the replay. People sign up, get reminders, join from the same URL and come back to it later for the recording. Attendees don’t need to download anything or create an account.

For marketers this simplifies promotion. You share one URL in emails, ads and social posts, and it never goes stale after the event.

Step 2: Capture Leads, Not Just Names

The registration form is where the webinar pays for itself. Restream collects registrations on the webinar page and lets you export webinar data as CSV, so you can push it into your CRM or marketing automation tool. Before you launch, decide which two or three fields will actually change how sales follows up, such as company size or current tool, and keep the form short.

Step 3: Produce It Like a Show

A webinar with a static slide and a talking head loses people after 10 minutes. Browser-based production closes that gap. In Restream‘s studio a host can:

• Switch between scenes and layouts during the session
• Add brand colors, logos and overlays
• Bring guest speakers on stage by sending them a link (up to 10 people in the studio, hosts and co-producers included)
• Invite co-producers to run the show as a team: one prepares guests in the green room, another manages the chat, links and calls to action
• Share screens and slides

Sessions can be produced in up to 4K, which matters if the recording will be reused in paid campaigns or on YouTube.

Step 4: Go Where Your Audience Already Is

Not everyone in your target audience will register. Some will only see you if you show up in their feed. Restream started as a multistreaming tool, so the same account can also take a live session to YouTube, LinkedIn or Facebook when you want a public audience. For internal or partner sessions, you can also stream into Slack.

Your speakers can amplify the session too. Guests and panelists can connect their own LinkedIn, X or YouTube channels, so the webinar also reaches their followers, broadcast securely through their own accounts.

Step 5: Turn the Recording into Ten Assets

This is where most teams stop and where the engine actually starts. After the session, Restream keeps a cloud recording, a transcript, and AI-generated clips with captions. Here is a simple repurposing plan for a 45-minute session:

Step 6: Measure What Matters

Webinar reporting usually stops at “registrants vs. attendees.” Add three more numbers: average watch time, replay views in the first 14 days, and pipeline influenced by registrants. Restream‘s analytics cover the viewing side, and the CSV export makes it easy to join with CRM data for the pipeline side.

Where It Fits in the Martech Stack

Restream is not a full marketing automation suite, and it doesn’t try to be. It sits between your CRM or email tool and your social channels. It is a good fit for teams that want broadcast-quality sessions from the browser and need to reach audiences beyond their own list. Teams that need complex multi-session virtual conferences with booths and networking lounges will still want a dedicated event platform.

The Takeaway

The webinar itself is the most expensive part of the content. Everything after it is cheap, as long as your tool makes recording, clipping, and exporting easy. Plan the repurposing before you plan the slides, and one hour live can carry your content calendar for weeks.

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: One Live Session, a Month of Content: Running Webinars as a Marketing Engine with Restream

On Tuesday, a developer gets a message from her manager:

Can we add a chatbot to the site by Friday?

She has one running in an hour. It looks great. It greets visitors, it answers questions, and it has a friendly name.

By Monday, it has told a visitor the company offers free international shipping. It does not. Nobody wrote that down anywhere. The chatbot just thought it sounded helpful.

That is the gap most tutorials skip. Getting a chatbot onto a page takes minutes. Getting one to answer correctly, from your own information, is the actual job, and it is what a good AI chatbot builder is for.

Why Most Business Chatbots Disappoint

Most guides on how to build an AI chatbot start with choosing a platform. We think that is the wrong end of the problem.

The Chatbot Was Not the Problem. The Content Was.

A chatbot can only be as good as the information behind it. If your shipping page is vague, the chatbot will be vague. If your returns policy lives in a PDF nobody uploaded, the chatbot has never heard of it.

So when a chatbot gives a bad answer, the first question is not Which model is it using? It is What did we actually give it to read?

General Models Guess. Your Visitors Notice.

A chatbot running on a general model has read a great deal of the internet and none of your website. Ask it about your delivery times and it will produce a confident, plausible answer based on what companies like yours usually do.

That is fine for a demo. It is not fine for a customer who is about to place an order based on what it said.

Speed Is Not the Same as Ready

Every tutorial on how to make a chatbot promises it will take minutes, and technically they are right. The widget appears, the greeting works, and the demo looks convincing.

What they rarely mention is that minutes is how long it takes to install the chatbot, not how long it takes to make it trustworthy. Those are different jobs, and only one of them is quick.

What an AI Chatbot Builder Actually Does

Strip away the marketing and a builder does three jobs: it reads your content, it answers from it, and it knows when to stop.

It Reads Your Pages Instead of the Internet

The core idea is simple. You train chatbot on your own data, meaning your pages, documents, and policies, rather than letting it answer from general knowledge. An AI chatbot builder crawls your site or takes the files you upload, and every answer comes from that material.

The best ones also show where each answer came from. That matters more than it sounds. When an answer is wrong, the source tells you exactly which page to fix.

What “Training” Actually Means Here

The word training causes a lot of confusion. For a chatbot on a business website, you are usually not retraining a model at all. You are giving it a library of your own content and telling it to answer only from that library.

That distinction matters when you plan how to make an AI chatbot for a business site. You do not need a data science team. You need accurate pages, a clear list of what the chatbot should and should not handle, and someone who checks the answers.

Chatbot Code vs a No Code Chatbot Builder

Developers reading this are probably thinking they could build one themselves. You could, and plenty of guides on how to build a chatbot from scratch will walk you through it. Writing your own gives you full control over retrieval, prompts, and behavior, and it is a genuinely interesting project.

It also means you own the maintenance, the hosting, the channel integrations, and every edge case forever. A builder trades some of that control for speed and a lot less upkeep. For most business websites, that is a trade worth making. For a product where the chatbot is the product, it might not be.

What It Should Never Do on Its Own

Some questions should never get an automated answer. Refund exceptions. Legal questions. Anything involving money that the chatbot cannot confirm from your own pages.

A good builder lets you draw that line clearly. The chatbot answers what it can support and hands everything else to a person, rather than trying to be helpful at your expense.

How to Create a Chatbot That Answers from Your Own Pages

Search for how to create ai chatbot tools and you will find plenty of five-minute tutorials. This is the slightly longer version of how to create a chatbot, the one that still works a month later.

Step 1: Write Down the 20 Questions Visitors Actually Ask

Before you touch any tool, collect the questions people already send you. Support emails, contact forms, sales calls, DMs. Twenty is enough to see the pattern, and the pattern is usually the same handful of questions asked in different words.

This list becomes your test script later. It is also the most useful thing in this whole process, and it costs nothing.

Step 2: Make Sure Your Pages Actually Answer Them

Go through the list and find the page that answers each question. Some will have one. Some will have a page that sort of answers it. Some will have nothing at all.

Fix the gaps now. A chatbot cannot answer from a page that does not exist, and it will not magically clarify a page that is vague.

Step 3: Load the Content into the Builder

Point the AI chatbot builder at your site, upload the documents that are not on the site, and paste in anything that lives only in someone’s head. Shipping rules, returns, pricing explanations, and product details are usually the first priorities.

Step 4: Test It with the Questions That Should Fail

Ask it your 20 questions and check every answer against the source. Then do the part most people skip: ask it questions your content does not cover.

The right answer to those is some version of “I do not know, let me get someone.” If it answers them confidently anyway, you have found the most important problem before your visitors did.

Step 5: Install It Where Visitors Already Are

A website chatbot is the obvious starting point, usually through a script tag or a plugin. But plenty of questions never reach the website. They arrive on WhatsApp, Instagram, or Messenger.

If you can, run one knowledge base across all of them. Otherwise you will end up with an AI chatbot for website visitors that says one thing and a messaging setup that says another.

Step 6: Decide When It Hands Off to a Person

Set the rules before launch. Which topics always go to a person. What happens when the chatbot does not know. How the conversation reaches your team.

The handoff should carry the whole transcript. Nobody enjoys explaining their problem twice, and the person taking over should not have to ask.

Step 7: Read the Conversations Every Week

The questions people ask are free research. The ones the chatbot could not answer are a list of pages you need to write. Ten minutes a week with the conversation log will do more for the chatbot than any setting you can change. It is also the easiest way to spot a new product question before it turns into a dozen support emails.

Choosing a No Code Chatbot Builder

There is no shortage of options, and most of them look similar on a pricing page. Whether you want a simple chatbot for website support questions or something that also handles sales conversations, the differences only show up when you test them properly, with your own content and your own awkward questions.

What to Check Before You Commit

Be skeptical of big claims. The US Federal Trade Commission has warned businesses to keep their AI claims in check, meaning they should not exaggerate what an AI product can do and should be able to back up what they say. The same caution applies when you are the buyer.

So test instead of trusting the demo. Load your own content, ask your own 20 questions, and ask the ones that should fail. Check that answers show their source, that it declines when it should, and that it works on the channels your customers use.

If the main use is AI chatbot customer service, test the handoff too. Start a conversation, ask for something only a person can do, and see what your team actually receives.

It is also worth asking what happens when your content changes. A custom AI chatbot that needs retraining every time you edit a page will fall out of date quickly. The good ones pick up changes to your documents without a project.

Questions to Ask Any Vendor

A short list goes a long way. What happens when the answer is not in our content? Can we see the source behind every reply? How does a conversation reach our team? What does it cost when volume spikes?

Any chatbot builder worth paying for will answer those plainly. If the reply to the first question is a long pause followed by “it is very accurate,” keep looking. An AI chatbot builder should be able to show you what it does when it does not know, not just what it does when it does.

How Agentency Handles an AI Chatbot Builder

We build one of these, so it is only fair to hold our own platform to the same test.

It Answers Only from What You Give It

Agentency answers from the material you provide: crawled pages, uploaded PDFs, or pasted text. Every answer shows the source it came from, and the built-in test chat lets you ask your 20 questions before any visitor does.

It Declines When the Answer Is Not There

When the retrieval quality gate finds nothing relevant, the reply is blocked. The agent says it does not know and hands the conversation to a person with the full transcript attached, so nobody has to repeat themselves.

It Installs Without an Integration Project

You add it with one script tag or the WordPress or Shopify plugin. No integration project, no model training, no data team. Training on your content runs in the background while you get on with something else.

One Agent Across 10+ Channels

The same agent answers on the website widget, WhatsApp, Instagram, Messenger, Telegram, Slack, and more, all from one knowledge base. It replies in the language the visitor writes in, so one setup covers visitors from more than one country.

It Checks Live Data and Asks Before Risky Actions

When an answer needs live information, like an order status, a Call Action lets the agent query an API you already run. Higher risk actions can require the visitor’s explicit confirmation before anything happens.

Behind all of it, data is encrypted in transit, each workspace is isolated, and two-factor authentication is available on every plan.

Frequently Asked Questions

What This Comes Down To

Anyone can get a chatbot onto a website by Friday. The hard part was never the install.

The chatbot that works is the one that answers from your own content, shows where each answer came from, and admits when it does not know. Pick an AI chatbot builder that does those three things, spend an afternoon on the 20 questions, and you will avoid the free international shipping conversation entirely.

Most of the work, in the end, is writing things down properly. The chatbot just makes it obvious when you have not.

Agentency is an AI chatbot builder for customer service and sales. Businesses train the Agentency AI agent on their own website, documents, and product catalogue, and it answers customer questions with the source shown. When the answer is not there, it declines and hands off to a person with the full transcript. One agent covers the website, WhatsApp, Instagram, Messenger, Telegram, Slack, and 10+ channels, in the customer’s language, for ecommerce, real estate, restaurants, and support teams.

©2026 DK New Media, LLC, All rights reserved | Disclosure

Originally Published on Martech Zone: How to Build an AI Chatbot for Your Website