Share of Model Is Not a Metric Yet: What Marketing Leaders Should Actually Measure in AI Search
I sat in a client review earlier this year where two AI visibility dashboards, both paid for by the same organisation, reported the brand's share of voice in ChatGPT eleven points apart. Nobody in the room could explain why. The head of digital asked which number she should put in the board pack. I told her neither, and that the honest answer was a range with a confidence note attached. That went down about as well as you would expect.
Thirty years in this industry and I have never seen a measurement category grow this fast on foundations this thin. The IAB now counts more than twenty companies selling AI visibility measurement, each with its own methodology, each capable of producing a different answer for the same brand. Only sixteen percent of brands systematically track AI visibility at all. Meanwhile Fractl data reported by Digiday puts roughly twenty four percent of search and content budgets into AI visibility work. We are spending like the measurement is solved. It is not.
What still holds up, for the forceable
Let me set the table before I clear it. Branded search volume is now one of the most commercially useful signals a brand has, because it is where AI influence surfaces. Similarweb work and our own at WPP Media found that roughly fifty six percent of AI influenced traffic arrives as branded search rather than an AI referral click, days after the conversation happened. Impression share in Google Search Console still matters, arguably more than clicks do. Paid search conversion and quality score still matter. Crawlability, structured data and site speed still matter, because a page a model cannot read is a page it cannot cite.
What has stopped working is the click as the primary unit of account. Ahrefs measured a fifty eight percent reduction in position one click through rate when an AI Overview appears, and Pew's browsing data found users clicked eight percent of the time with an AI Overview present against fifteen percent without. If clicks are still your headline KPI, and Kantar research suggests they are for around seventy eight percent of brands, you are measuring the shadow rather than the object.
What actually changed, and why it matters now
Three things happened in quick succession this year that should reset how CMOs think about this.
First, Google gave us a real, if partial, measurement surface. On 3 June 2026 Google launched dedicated Search Generative AI performance reports in Search Console, rolling out to UK site owners first under CMA pressure. For the first time you can separate visibility inside AI Overviews and AI Mode from ordinary organic. The catch is significant: impressions only. No clicks, no click through rate, no queries. It answers "did I appear" and refuses to answer "what was it worth".
Second, the scale question stopped being theoretical. Semrush analysed 126 million US AI search prompts between January and April 2026 across ChatGPT, Gemini, AI Mode and AI Overviews, covering 1,200 brands in 22 industries. Only 36 of those brands held a top 100 position on every platform in every month. The same study found ChatGPT citing an average of fifteen sources per response against Gemini's three. Per platform measurement is not a nice to have. A blended visibility score across engines that behave that differently is an average of incompatible things.
Third, and this is the one that should worry vendors, Rand Fishkin ran 2,961 controlled tests across ChatGPT, Claude and Google AI with 600 volunteers. Ask an AI for brand recommendations a hundred times and you get the same list fewer than one time in a hundred, and the same list in the same order fewer than one time in a thousand. His conclusion, which I share, is that visibility percentage across many prompts run many times is defensible, and any tool selling you a ranking position in AI is selling you nothing.
The frameworks I would build on, and where I would push back
The IAB's Measuring Visibility in the AI Era is the most useful thing published on this in eighteen months. Its four P's give us a causal hierarchy rather than a metrics soup:
- Presence, whether you appear at all
- Prominence, where and how visibly
- Portrayal, in what context and with what accuracy
- Persuasion, whether any of it moves anything
Its real contribution is the split between directional and decision grade data, and a floor that should embarrass a lot of vendor decks: fewer than fifty queries in a measurement programme is classed as exploratory, not even directional.
Two honest caveats. Caroline Giegerich told AdExchanger the IAB deliberately avoided calling this a standard, because standards need stability and the market has none. And the framework measures single query, single response visibility. That is not how people use these tools. Real decisions unfold over four, six, ten turns of a conversation, and a brand can be named confidently at turn one and gone by turn five.
That gap is why I use the PLUS framework with clients alongside the IAB vocabulary. Four factors the IAB does not fully price in. Personalisation: responses adapt to account history, so universal rankings are a fiction and you should audit by persona instead. Location: outputs are hyper localised, so measure local citation accuracy across your top territories. Uniqueness: models are probabilistic, so batch test the same prompt fifty times and report mention probability, not a snapshot. Stateful ness: context carries, so track brand decay rate across a thread. I want a brand holding above fifty percent mention visibility at prompt four and above forty percent by prompt six. That number tells me more about commercial resilience than any single answer share ever will.
Traditional metrics and their AI era equivalents
| Traditional metric | AI era equivalent | Why it matters |
|---|---|---|
| Keyword ranking position | Mention probability across 50+ runs of the same prompt | Position in an AI answer is close to random; frequency across repeated runs is the only stable signal |
| Share of voice | Share of model, segmented by platform and persona | ChatGPT and Gemini cite radically different source volumes, so a blended score hides the diagnosis |
| Organic click through rate | Citation rate plus post citation click through rate | You can be consumed without being clicked; citation is the exposure event, the click is now optional |
| Impressions | Presence rate and visibility momentum | Search Console generative AI impressions confirm appearance, momentum tells you whether you are gaining or drifting |
| Brand sentiment tracking | Portrayal: sentiment, framing, hallucination rate, factual inaccuracy rate | Being framed as "the budget option" is visibility that quietly erodes pricing power |
| Last click attribution | Branded search lift plus assisted conversion, corroborated by media mix modelling | Most AI influence lands days later as branded search, invisible to referral reporting |
| Single query snapshot | Brand decay rate across a conversation thread | Purchase decisions happen at turn six, not turn one |
How I would actually track this
The approachI recommend is three layered, and no single vendor covers it. Platform native data first: Search Console generative AI reports, and GA4 with a properly defined AI assistant channel, because a meaningful share of AI referrals still lands in direct if you leave the defaults alone. Then an active query simulation tool, Profound, Peec, Semrush's AI Visibility Toolkit, AirOps, Scrunch, whichever survives procurement, used for trend and competitive context rather than truth. Then panel or clickstream data to see what real people ask, because synthetic prompt libraries encode the vendor's assumptions about your category.
On procurement, use the IAB disclosure list as an RFP instrument. Ask for platform coverage with model versions, prompt library construction and refresh cadence, query sourcing, data collection architecture, and how re baselining is handled after model updates. If a vendor will not answer, that silence is your answer.
Does AI visibility replace SEO rankings? No, and asking is the wrong question
Rankings have not become worthless, they have become an input rather than an outcome. Citation overlap between AI Overviews and traditional top ten results has fallen sharply, from roughly three quarters in mid 2025 to somewhere between a fifth and a half by early 2026, depending on whose methodology you trust. That range is itself the story. You now need to rank and be citable, and those are different jobs.
The ROI case, made properly
This is where most AI visibility business cases fall over, and where I would push back hardest on the vendor narrative. Adobe found AI referred visitors to US retail sites converting fifty four percent better than non AI traffic in May 2026, across more than a trillion visits, a reversal from converting worse a year earlier. Contentsquare, measuring a different thing, put AI referred visits at around 0.2 percent of all visits. Both are sound. The reconciliation is the insight: tiny volume, exceptional quality, steep trajectory.
So do not build the case on referral traffic. Build it on influence. WPP's research with Oxford's Saïd Business School found that eighty four percent of purchases go to brands consumers were already predisposed toward before the buying moment. AI answers are now a predisposition engine. Digiday reported CMOs struggling to connect visibility to sales, with the more sophisticated teams triangulating visibility tools, paid conversion data, branded search movement and their own media mix models rather than waiting for one dashboard to explain everything. That triangulation is the answer. There is no single tool that paints the whole picture, and anyone claiming otherwise should be shown the door.
What I would do in the next ninety days
Days one to thirty: establish a variability baseline before you report a single number upward. Run your priority prompts repeatedly and document what normal fluctuation looks like, so you can tell a model update from a competitor win. Anything below fifty queries, do not report it.
Days thirty one to sixty: rewrite your reporting language. Report ranges, not points. Report per platform, not blended. Separate platform driven shifts from market driven ones explicitly, and tell your board which is which. If your organisation is UK based, check whether the Search Console generative AI report has reached your property and start the baseline now, because there is no historical backfill.
Days sixty one to ninety: connect visibility to a commercial signal. Branded search lift is the cheapest and most honest bridge available. Then take the IAB disclosure requirements into your next vendor review and let the ones who cannot answer them fall away.
The uncomfortable truth is that we are early, and the industry is pricing this as though it is late. Measure carefully, report humbly, and be the person in the room who knows the difference between a number and a vibe with a decimal point on it.
Sources and further reading
- IAB, Measuring Visibility in the AI Era, 3 August 2026: https://www.iab.com/guidelines/measuring-visibility-in-the-ai-era/
- IAB press release on the four Ps and the decision grade standard: https://www.iab.com/news/iab-releases-measuring-visibility-in-the-ai-era/
- AdExchanger, on why the IAB avoided the words standard and framework: https://www.adexchanger.com/ai/iabs-new-advice-on-how-to-measure-ai-search-visibility/
- PPC Land, on the sixteen percent tracking figure: https://ppc.land/only-16-of-brands-track-ai-visibility-as-iab-sets-measurement-standard/
- SparkToro and Gumshoe, AI recommendation consistency research, 2,961 tests: https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/
- Semrush, 2026 AI Visibility Index, 126 million prompts: https://www.semrush.com/news/463141-semrush-releases-expanded-2026-ai-visibility-index-analyzing-126-million-ai-search-prompts/
- Google Search Central, Search Generative AI performance reports, 3 June 2026: https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports
- Digiday, CMOs struggling to link AI visibility with sales: https://digiday.com/marketing/cmos-are-struggling-to-link-ai-visibility-with-sales/
- Digiday, marketers questioning AI visibility tool value: https://digiday.com/marketing/marketers-question-expensive-ai-visibility-tools-as-inconsistent-results-fuel-skepticism/
- Digital Commerce 360 on Adobe Analytics AI referral data, June 2026: https://www.digitalcommerce360.com/2026/06/17/adobe-ai-referred-traffic-to-retail-sites-doubles-in-a-year/
- Forrester, on the B2B visibility vacuum: https://www.forrester.com/blogs/is-ai-visibility-your-2026-imperative-learn-how-to-achieve-it-at-b2b-summit
- AIVO Journal critique of the IAB framework's single query assumption: https://www.aivojournal.org/measuring-the-wrong-era-why-the-iabs-new-visibility-framework-cant-see-past-the-query/
| Number | First Name | Last Name | Email Address |
|---|---|---|---|
| 1 | Anne | Evans | anne.evans@mail.com |
| 2 | Bill | Fernandez | bill.fernandez@mail.com |
| 3 | Candice | Gates | candice.gates@mail.com |
| 4 | Dave | Hill | dave.hill@mail.com |









