Ask a marketing leader where their brand ranks on Google for its main category term and you will usually get a number, or at least a dashboard that has one.
Ask what share of AI answers name their brand when a buyer asks for a recommendation in that same category, and the conversation changes. Not because anyone is being careless. The number generally does not exist anywhere in the organisation.
That gap has now been measured, and it is wider than most teams assume.
The blind spot is the norm
Semrush published its 2026 AI Visibility Index on 26 June 2026. Two findings from its survey component:
- 45% of marketing leaders, in Semrush's words, "cannot accurately measure their brand visibility within AI-generated answers."
- Only 9% have tooling that covers every relevant metric across platforms.
Worth being precise about what those numbers are, because this is where most coverage of this study gets sloppy. Semrush ran two things: an analysis of 126 million US AI search prompts from January to April 2026 across ChatGPT, Gemini, Google AI Mode and Google AI Overviews, and a separate survey of marketing organisations. The 45% and the 9% come from the survey, and the respondent count is not stated on the release. They are not derived from the 126 million prompts. Anyone quoting them as if they were is describing a different study than the one that exists.
With that caveat, the finding stands and it is useful. If you cannot answer the AI-share question for your own brand, you are with the majority, not behind it. The 9% are the outlier.
What the measurement is missing
There is a second trap underneath the first. Even teams that do start measuring tend to measure the surface they control, their own website, and that is not where the citations are. In the June 2026 Cited Index, only 4.1% of brand citations point back to the brand's own site; the other 95.9% sit on third-party domains the brand cannot edit. Muck Rack's May 2026 study of 25 million-plus links found the same shape from a different angle, with earned media at 84% of AI citations and paid content at 0.3%. Watching your own site tells you almost nothing about where you actually appear.
Nearly half our sample is missing somewhere
Want to know how your brand scores on these same metrics?
We'll run 20 prompts across 3 AI platforms and send your report within 24 hours.
Here is the finding that changed how I talk about this in audit calls.
Across the 226 brands in the June 2026 Cited Index, 108 are completely invisible on at least one of the five engines. Not scoring low. Scoring zero. Only 118 brands, 52.2% of the set, register on all five.
| Engine | Average brand visibility | Brands at zero visibility |
|---|---|---|
| ChatGPT | 14.5 | 30 of 226 |
| Google AI Mode | 12.2 | 27 of 226 |
| Gemini | 11.3 | 30 of 226 |
| Google AI Overviews | 11.0 | 37 of 226 |
| Perplexity | 9.5 | 45 of 226 |
Source: June 2026 Cited Index, 226 brands, 270 prompts, 1,278 responses. Methodology.
Here is the trap in that table, and it is the real finding. The averages sit close together, 9.5 to 14.5, which makes the five engines look roughly interchangeable, as if measuring one is a fair proxy for the rest. They are not. The brands sitting at zero are different brands on each engine. A company can be perfectly healthy on ChatGPT and structurally absent from Perplexity, and neither the average nor a single-engine check will ever show it. We documented the same divergence at brand level in platform bias, where one brand scores near the top on one engine and near zero on another in the very same week. Even Google's own two surfaces, AI Overviews and AI Mode, disagree about the same brand.
So the 108 is not really a story about 108 brands. It is a story about the measurement itself. If you check one engine and see your brand, you have learned that you are present on that engine and nothing more. The absence lives on the surface you did not check.
Three numbers worth asking for
If the measurement layer is new to your team, these are the three I would start with, in this order.
How many of the five engines name you at all. This is a binary and it takes the least work to establish. It catches the failure mode that averages hide, which is total absence from one surface. Nearly half of the brands we track fail this check somewhere.
Your share of answers inside your own category. Not visibility in general. Category-level buying prompts, phrased the way buyers phrase them, with budgets and local constraints included, are where shortlists actually form. Aggregate visibility across unrelated prompts is a vanity number.
The descriptors that come with your name. When an engine does name you, what adjective rides alongside? In our CRM data the engines attach affordable and customizable to one brand and enterprise and expensive to another. Both appear in the answers. Only one survives a budget-framed shortlist.
None of those three require buying anything to begin.
Here is the version you can run yourself this week. Write five prompts the way a buyer in your category would actually type them, not the way a marketer would. Include the constraints real buyers include: a rupee budget, a team size, an integration they already use, a city. "Good CRM for a 15-person sales team in Pune under ₹800 per user" is a real prompt. "Best CRM software" is not, and it will return a generic global answer that tells you nothing about your position.
Run all five against all five engines. Log three columns per run: whether your brand is named, roughly where in the answer it sits, and which words the engine used to describe you. Twenty-five cells, one afternoon.
The column that usually surprises people is the third one. Absence is easy to accept once you see it. Being named with the wrong adjective, by an engine your buyer trusts, in an answer you never saw, is the part that tends to change the budget conversation.
That single exercise moves a team out of the 45% faster than any dashboard purchase.
What I do not yet know, and would like to, is whether the brands sitting at zero on one engine got there through something fixable or through absence from the third-party sources that engine happens to favour. Our dataset shows the outcome, not the cause.
What is not uncertain is the cost of not checking. A brand at zero on one engine gets no warning and no bounce report. The buyers who built their shortlist out of that engine's answer never read the brand's name in it, and never had a reason to look further. Nothing in your analytics will show you a deal you were never in. If the pattern holds in the August edition, 108 is the number I will be watching, because it says this problem is binary before it is incremental.