Measuring SEM when AI platforms hide the data

As of November 2026, the honest answer to "how do you measure digital marketing" is that you cannot measure every influence, so you measure what you can defend. The IAB's AI-Era Attribution Blueprint, published on 12 November 2026 after a working group that included Walmart, WPP Media, Criteo, Microsoft and Prudential Financial, says as much: conversions shaped by an AI recommendation often land in search or direct traffic, and the data that would prove the AI did it sits inside platforms advertisers cannot read. When you hire a sem agency, the first question to ask is where its attribution starts, before any talk of dashboards.

Four terms carry the rest of the discussion, and each one is defined here so that a later disagreement can point at a definition rather than a feeling.

• Attributed conversion: a lead or sale a report assigns to a channel through a defined rule. It counts what the rule can see.
• Influenced conversion: a sale an AI answer contributed to that no report assigns to that answer. The IAB puts these in a third category, "contributory" claims, and asks what evidence would support one.
• Visibility metric: how often a brand appears, gets described accurately or gets cited in AI answers. It says nothing about revenue on its own.
• Signed-deal ratio: revenue from closed deals divided by ad spend over one fixed period. It survives a change of platform because both sides of the fraction are yours.

The IAB could not finish the standard, and its own report says why

Digiday reported on 12 November 2026 that the IAB spent the effort over four months and ended with a framework that reads more like a call to action than a rulebook. Caroline Giegerich, the IAB's vp of AI, told the roundtable that there are evidence layers missing and that the work started in one place and ended in another.

That outcome is worth understanding before you judge any vendor claiming an "AI attribution" capability. The blueprint classifies three pathways to a purchase: human-owned, AI-assisted and AI-executed. It then asks who can claim credit for each and what evidence the claim needs. The answer in the document itself is that the evidence is often not available.

What the framework publishes is a map of what is observable:

The last row is the one that keeps a credit claim from being substantiated. Digiday quoted the IAB's own text on the publisher case: a review shapes a sale, the retailer records a search or direct conversion, and the connection between the sourcing signal and the sale is missing.

How much AI-influenced traffic is reported as search or direct

Similarweb data cited in the IAB report gives the size of the reporting error. Consumers who received a brand recommendation from an AI assistant were 2.5 times more likely to visit that brand's website within seven days. Of those visits, only 8.8% came directly from AI referrals. Search took 55.9% and direct traffic 19.9%.

Read those figures with the definition in mind. They describe visit attribution, not sales attribution, and they describe a seven-day window. An advertiser looking at the same campaign in Google Ads or GA4 would see search doing well and would have no reason to doubt it. The number is not wrong; it is answering a narrower question than the one the budget asks.

Profound's browser-session analysis, also reported by Digiday, puts the consumer side at one in three consumers having a relevant AI conversation about their purchase within a month that went unattributed. That dataset excluded Google's AI Mode and AI Overviews, so it is a floor for the miss, not a ceiling.

How to measure digital marketing when the platform data stops

This is the gap the search results leave open, and it is the section to keep if you keep only one.

The mistake is to look for a better platform report. No platform report can supply what the IAB says is missing, because the missing part happens inside someone else's system. What you can build is a ledger with a fixed definition on each side, owned by you, that does not depend on any one platform's view of the world. That ledger is also the strongest test of an sem agency brief, because it fixes what the agency is paid to move.

1. Fix the metric and the period before the campaign starts. Pick one ratio, such as closed-deal revenue divided by ad spend, and one reporting period. Write both into the agency brief.
2. Take the starting point from your own data layer. Since the platform's reporting cannot see the gap, the ledger starts from your CRM, your web analytics and your click identifiers. This is the point where the platform's report becomes evidence to be checked, not the answer.
3. Keep paid and organic on one page. If search absorbs the AI-influenced visits, an organic and paid split will disagree with any single-channel view. One ledger with both rows shows the disagreement instead of hiding it.
4. Separate what you can claim from what you can only observe. AI visibility metrics, such as the share of tracked answers that mention the brand, belong in a separate layer with its own limits stated. Do not add them to a revenue number.
5. Report every period, including the ones where the ratio drops. A ledger that only appears in good months is a marketing document.

A worked example makes the difference clear. If search is credited with 55.9% of AI-influenced visits, a search campaign looks responsible for a lead its budget did not produce. Move the same leads onto a ledger that starts at the CRM record, and the question changes from "which channel got the last click" to "which spend produced the closed deal". The second question is the one that survives an argument with finance, and it is also the one a well-run sem agency will volunteer before you ask.

What the platform record cannot see: postback windows and blind spots

Platform reports also fail for reasons that have nothing to do with AI, and these are easier to test.

The first is the attribution window. A platform accepts a conversion only within its own window and rejects anything outside it. Meta's window runs seven days and Google's runs up to a month. A B2B deal that closes in the second month never enters either model, so the platform optimises against a shortened version of the funnel.

The second is scope. No platform's algorithm allocates budget across other platforms. Google Smart Bidding allocates inside Google; Meta's automated rules stay inside its own campaign structure. Each platform reports on the part of the journey it owns.

The third is the space outside the ad account: in-site interaction, form completion, lead progression and CRM follow-up. Platform reporting does not carry these, and that is where a funnel usually loses the budget it spent.

What Topkee measures instead of a platform-reported conversion

This is where the vendor view gets specific, and it rests on what Topkee publishes about its own reporting.

Topkee builds its Campaign Intelligence Report on one report fed by four data sources: Google Ads, Yahoo Ads, Facebook Ads and GA4, with the GA4 row as the entry point for organic traffic. The attribution starting point is the client's own data layer rather than a platform backend, which follows the same logic the IAB found missing and reverses the direction of the question: the report asks what the platform's numbers look like under one standard. The fixed metric is signed-deal value divided by ad spend, delivered every period, so a change in channel mix shows up in one number rather than in two dashboards that disagree. For an advertiser already running an sem agency account, the report is a way to test the existing structure rather than replace it.

The stitching layer is Topkee's TTO CDP. Each user gets a TTCLID that joins Google Ads interactions to the rest of their touchpoints, and qualified conversions post back to Google and Meta so smart bidding learns from revenue rather than clicks. The data sits in the client's own tenant. Topkee Media, founded in Hong Kong and operating in Singapore, sells this as part of its Performance Ads subscription, delivered on accounts the client already runs.

On the AI side, Topkee's GEO service measures whether the brand gets mentioned, described accurately and cited in generative answers, with formulas kept fixed: mention rate is valid answers containing a brand mention divided by valid answers under identical conditions. Its documented limits matter more than the metric. Samples come from agreed API environments and are not what every consumer sees, and the citation roll-up cannot be called a visible citation rate. Topkee states that AI exposure cannot yet be attributed to revenue, which is the same gap the IAB describes.

What to ask a sem agency singapore buyer can trust

The cluster of questions underneath this one is what an SEM agency does with measurement, and Singapore buyers can test the answer in five questions.

1. Where does your attribution start: the ad platform's report or the client's own data layer?
2. What is the one fixed ratio you report every month, and how is it calculated?
3. How do conversions outside the platform's postback window get into the number?
4. Which metric in your AI visibility work is an influence measure rather than a traffic measure?
5. When the data is missing, what do you report instead of a number?

An agency that answers the first question with a dashboard name has not yet been asked the question properly. Singapore buyers choosing a sem agency singapore option usually compare account structures and case studies; the ledger question ranks higher because it decides what the case study can prove. Marketing Singapore: What a Media Agency of Record Takes On sets out the wider scope of that relationship, and Prove AI Visibility ROI: Metrics a CFO Will Accept goes deeper on the AI-side metrics and how to present them to a finance team that will not accept a platform screenshot.

What may change the picture

Two developments in the news item point at a smaller gap rather than a solved one. The U.K.'s Competition and Markets Authority requires Google to give U.K. publishers AI-search clickthrough data by December 3, and OpenAI is testing a dashboard that counts how often a business appears in answers and how many clicks those appearances send.

Neither closes the loop to the sale. Click data from an AI answer gives you a visit, not a closed deal, and publisher click data addresses a different reader from the advertiser buying search. Treat both as an input to the ledger, and keep the ledger's definition unchanged when they arrive. Any sem agency that rebuilds its attribution model around the new data, rather than around a better story about the old numbers, is the one worth keeping.

FAQ

How do I measure digital marketing when AI search hides the conversion?

Measure the part you own. Fix one ratio, such as closed-deal revenue divided by ad spend, over a fixed period, and start it from your own data layer rather than a platform report. Record AI visibility metrics separately as influence measures, and state their sampling limits.

Can I attribute a sale to an AI recommendation yet?

Not fully. The IAB's blueprint sets out the evidence a credit claim would need and finds that the link between sourcing signals and downstream conversions is largely missing. Publishers may see when their content is cited, but citation alone does not show how much it influenced the answer.

What does the IAB's AI-Era Attribution Blueprint actually contain?

It classifies AI's role in a purchase into human-owned, AI-assisted and AI-executed pathways, then sets out who can claim credit and what evidence each claim needs. It also documents the transparency and measurement gaps that make those claims hard to verify.

Why do AI-influenced visits show up as search or direct traffic?

Because the AI platform's referral often is not passed through, so the visit lands with whatever referrer the browser or app supplies. Similarweb's data cited in the report put only 8.8% of those visits down to AI referrals, with search and direct taking most of the rest.

What should I ask an SEM agency about AI measurement?

Ask where attribution starts, which fixed ratio is reported every month and how it is calculated, and how conversions outside the platform's postback window are handled. Ask which of its AI metrics measure influence rather than traffic.

Sources

1. Sara Guaglione, "Media Briefing: The IAB finds AI measurement standards elusive for publishers amid platform data gaps", Digiday, 2026-11-12. https://digiday.com/media/media-briefing-the-iab-finds-ai-measurement-standards-elusive-for-publishers-amid-platform-data-gaps/
2. "The AI-Era Attribution Blueprint", IAB, 2026-11-12. https://www.iab.com/guidelines/the-ai-era-attribution-blueprint/