What MFA sites actually are (and why the definition keeps shifting)
Made-for-advertising sites are properties built primarily to capture programmatic ad revenue rather than to serve readers. The content is incidental. The business model is the inventory stack: dense ad units, aggressive refresh logic, auto-play video, and traffic sourced through content recommendation widgets that deliver high impression counts at low reader attention.
The label has been contested since it emerged as an industry term. Some publishers with legitimate audiences run high ad density out of necessity; some MFA operators have learned to mimic editorial trappings convincingly. The structural signal matters more than the surface. A site whose revenue model only works when ads outnumber content, or whose traffic collapses without paid sourcing, is functionally MFA regardless of how the masthead reads.
Generative AI has sharpened the problem. Content that once required a room of low-cost writers can now be produced at near-zero marginal cost per page. A site that previously needed weeks to build topical depth for a given keyword cluster can produce hundreds of articles overnight. The barrier to launching a credible-looking MFA property has dropped close to zero, which means the scale of the problem grows with every model capability improvement.
Google's own publisher policies are explicit about the structural failure: ads served on pages where paid promotional material outnumbers publisher content are a policy violation. The same policies prohibit serving ads on "low-value content" and on pages with "embedded or copied content from others without additional commentary, curation, or otherwise adding value." Those policy lines describe the MFA playbook almost word for word. The gap between policy and enforcement is where the arbitrage lives.
How MFA inventory routes into brand budgets
Most brands running open exchange programmatic are not buying MFA inventory directly. They are buying audiences, keywords, and contextual categories through a DSP, and the DSP's algorithm finds the cheapest impression that matches the targeting parameters. MFA sites are structured to win those auctions: they produce high volumes of cheap impressions in broad content categories (news, lifestyle, finance, entertainment) that match common targeting setups.
The path from brand budget to MFA site typically runs through a supply-side platform and multiple intermediaries. Each hop adds a margin cut and reduces transparency. By the time the impression renders on the MFA property, the buying team's dashboard shows a category match, a viewability number (often technically accurate because the ad did render), and a CPM that looks efficient. What it does not show is that the page had seventeen other ads loading at the same time, the reader was driven there by a clickbait widget, and the session lasted under thirty seconds.
The efficiency signal is a trap. MFA inventory tends to show lower CPMs because the supply volume is enormous and the audience quality is poor. A campaign that achieves a low CPM by routing a large share of spend to MFA inventory is not an efficient campaign. It is an inefficient one that looks efficient on the metrics that get reported.
Industry bodies including the IAB Tech Lab and ANA have published guidance that classifies MFA inventory and calls for supply chain actors to flag or filter it. The IAB Tech Lab released a formal MFA Definition and Detection Framework in 2024; the ANA has published supply chain transparency studies and recommendations covering the same problem. Verification vendors DoubleVerify and IAS both operate MFA classification categories within their brand safety products. But classification operates at the domain or URL level, and MFA operators are faster at creating new domains than classifiers are at flagging them.
The structural fix is not faster classification. It is reducing the open exchange pool available to MFA operators in the first place, which is an inclusion-list problem rather than an exclusion-list problem. We return to that below.
Detection signals buyers can use
No single signal is definitive. MFA sites that have learned to game one detection vector often fail on another. The reliable approach is to evaluate signals in combination and to weight structural evidence over surface indicators.
Ads-to-content ratio. This is the defining structural signal. A page where ad units occupy more visual space than editorial content is flagging its own purpose. Counting ad slots versus word count is a crude proxy; a more useful measure is visual real estate. Pages with interstitials before content loads, ads above the fold with editorial content below, or sticky video units overlaying text are all high-ratio patterns. Google's publisher policy language about "more ads or other paid promotional material than publisher-content" describes the threshold with some precision.
Traffic sourcing. MFA sites typically cannot sustain traffic organically because their content is not genuinely useful. They rely on paid distribution: content recommendation widgets (the "around the web" grids at the bottom of news pages), social arbitrage, and occasionally bot-amplified traffic. Sellers.json and ads.txt provide some supply chain transparency, but they do not reveal traffic sourcing. Third-party tools that analyse referral patterns can surface anomalous traffic compositions. A site drawing the majority of its traffic from paid content widgets is a red flag regardless of the content quality.
Refresh and session behaviour. Auto-refreshing ad units inflate impression counts without corresponding attention. A user who lands on a page, decides it is not useful, and leaves within twenty seconds may generate multiple ad impressions through refresh before departure. Session duration and pages-per-session are not metrics most buyers see in their DSP reporting, but they are available through publisher-level analytics that supply path transparency initiatives are beginning to surface.
Ad slot density and placement patterns. More than five or six non-native ad placements on a standard article page is an elevated signal. Ads placed inside the content body at unusually frequent intervals (every paragraph, for instance) indicate a page designed to maximise visual contact rather than reading flow. Sticky video players that continue running after the reader has scrolled past them exploit viewability measurement without capturing genuine attention.
Content depth and originality. Generative AI has made thin content harder to identify by surface reading. An MFA page generated by a language model may be grammatically correct and topically relevant. The diagnostic is depth: does the content provide original analysis, cite primary sources, or demonstrate expertise that could only come from direct experience? Shallow content that covers the same keywords as a hundred other pages without adding anything is the editorial equivalent of the ads-to-content signal.
Supply path optimisation reduces MFA exposure by cutting indirect hops. Buyers who push their DSPs to shorten supply paths see fewer MFA impressions in reporting, not because MFA sites disappeared, but because those sites monetise primarily through aggregators and resellers, not direct publisher relationships.
Inventory quality checklist
Run this against any placement domain you are considering adding to an inclusion list or want to audit in your current open exchange mix. The checklist is not a fraud detection system. It is a structured review of structural signals.
How generative AI changed the MFA threat level
The MFA playbook predates large language models by years. Content farms in the early 2010s used low-cost outsourced writing to populate keyword-targeted sites with enough text to pass basic quality filters. What generative AI changed is the economics of scale and the speed of adaptation.
A competent MFA operator can now spin up a new domain, populate it with hundreds of contextually relevant articles, configure ad networks, and start bidding on supply within a few days. The content looks different from the machine-translated spam that characterised earlier content farms. It passes spelling and grammar checks. It achieves basic topical relevance. It does not pass an editorial quality test, but editorial quality is not what programmatic auctions measure.
The acceleration matters for buyers in three ways. Domain blocklists decay faster. A list of known MFA domains compiled last quarter is already partially stale because new domains have launched since. Exclusion-list strategies require continuous maintenance to remain effective. The similarity between AI-generated content on legitimate sites and AI-generated content on MFA sites also makes content quality harder to use as a standalone signal. Publishers with genuine editorial operations use generative AI to assist writers; the output can look superficially similar to MFA-generated text. And the volume of MFA inventory grows with model capability improvements, so the absolute amount of brand spend at risk increases even when percentages stay flat.
The connection to brand safety is real but specific. MFA sites are not brand-unsafe in the same sense that sites hosting harmful content are brand-unsafe. A financial services ad appearing on a thin lifestyle article is not a reputational incident. The harm is subtler: the ad delivered no genuine attention, consumed budget that could have reached a real audience, and contributed to a measurement environment where brands systematically overestimate campaign effectiveness. Reach and frequency metrics on campaigns with heavy MFA exposure are not accurate.
Leapbuzz runs a hard editorial gate on its own content for the same structural reason MFA operators exploit the lack of one. When content cannot survive reader scrutiny, it relies on algorithmic amplification. Content that earns genuine readership does not need the arbitrage. Sites that exist to generate ad impressions rather than serve readers are producing AI slop, and the distinction matters when your brand is paying for placement.
Inclusion-list thinking: why allowlists beat blocklists at scale
The dominant industry instinct is to fight MFA with exclusion lists: block known bad domains and keep the rest of the open exchange available. That is the wrong architecture for a world where new MFA domains launch continuously and AI-generated content makes surface-level detection unreliable. Inclusion-list thinking reverses the default: start with a defined set of verified publisher relationships and only expand when a new publisher passes review. The open exchange remains available for prospecting, but the bulk of impressions run against a curated pool.
In practice this means three structural moves. First, allocate a meaningful share of programmatic display budget to private marketplace deals with publishers whose audiences match your targeting and whose inventory you have reviewed manually. PMP deals bypass open exchange dynamics entirely; MFA inventory does not appear in PMP unless you build a PMP with an MFA operator, which requires active effort to avoid.
Second, ask your DSP to surface supply path information and prioritise direct publisher-SSP relationships over reseller chains. Supply path optimisation (SPO) reduces MFA exposure because MFA properties monetise primarily through aggregators, not direct SSP integrations. Third, apply the checklist above before adding any domain to an active inclusion list, and review the inclusion list on a quarterly cadence as new publisher data becomes available.
CTV and retail media sit outside the open web MFA pool. Inventory is smaller and more verifiable, publisher identity is clearer, and the arbitrage that dominates display does not transfer directly. The inclusion-list discipline matters most on open web display. For the CTV quality angle see the programmatic CTV partner guide; for closed-loop measurement see retail media CTV.
The measurement layer is where the inclusion-list shift produces the most visible results. When you move spend from open exchange to curated placements, reported CPMs usually rise. Apparent efficiency drops. But campaign outcome metrics, where they are measured properly, tend to improve or hold steady. A higher CPM against a real audience is a better investment than a lower CPM against a bot-amplified MFA property. Connecting your programmatic quality decisions to outcome measurement rather than impression-side efficiency metrics is the analytical move that makes the argument internally. See our media integration service for how we approach this measurement architecture with clients across Singapore, Australia, the US, Canada, and Malaysia.