AI search visibility for CMOs: eight questions to ask your team, and what good answers sound like

What AI search visibility means for a CMO, and the figures to know first

AI search visibility is how often, and how accurately, your brand, products and pages appear inside AI-generated answers, and whether those answers link back to you. For a chief marketing officer (CMO), it sits next to share of search as a measure of demand you can see before it reaches the funnel.

Googlebot held 39% of AI and search crawler traffic in July 2025, against 11.7% for GPTBot and 2.4% for Bytespider.

Share of AI and search crawler traffic, July 2025

Share of AI and search crawler traffic, July 2025

Label	Value
Googlebot	39%
GPTBot	11.7%
Meta crawler	7.5%
Amazonbot	5.9%
Bytespider	2.4%

The terms your team will use

| Generative engine optimization (GEO) | Our working term for the practice of making your content eligible, retrievable and citable in AI answers. |

The figures to know first

Cloudflare's August 2025 analysis gave the scale of the shift:

Those referral figures describe news sites, not your category. In our reading, the useful takeaway for a board is directional: more of your content is being read by machines, and fewer readers may follow a link back.

A first check, as a worked example

Take an example, not a benchmark: your team lists 50 priority pages and finds 8 that are not indexed or are blocked from showing a snippet. That leaves 42 pages eligible for AI features, and the 8 failures are a fix list with a clear owner. Run the same count each quarter and report it next to brand tracking results.

How Google's AI Overviews and AI Mode choose and link sources

Query fan-out, and what it does to your keyword list

A page can therefore be linked for a subquery the user never typed. Here is an example, not a benchmark. Your team picks 8 priority buying questions and writes down 20 likely fan-out queries for each, so 8 questions with 20 queries each gives 160 queries to check by hand for which pages appear.

In our reading, each of these changes rewards pages that Google can already rank and identify as the official source. That makes core SEO hygiene the first item on the CMO's checklist.

Which AI crawlers fetch for search and which collect training data

The split that matters for a CMO is simple. Some crawlers decide whether your pages can appear in an AI answer. Others collect content to train models. A third group fetches a page only when a user asks a question. Your team can block training without dropping out of search, but only if each token is set on purpose.

A robots.txt template to hand your team

This is our starting point for a company that wants AI search presence but no training use. Adjust paths to your site.

User-agent: Googlebot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: GPTBot
Disallow: /

Should we block AI crawlers like GPTBot

Treat crawler access as a governance decision with an owner, a written policy and a review date. Leaving it as a setting someone in IT changed once is how it goes wrong. Blocking bundles two separate choices: whether your content trains AI models, and whether it appears in AI answers. The previous section showed that OpenAI lets you make those choices independently and that changes take about a day to apply. Two further details matter for the policy.

Robots.txt is a request, not a lock

The practical point: a robots.txt line records your intent. Enforcement happens at the network or firewall layer, and that is a security budget conversation as much as a marketing one.

The trade-off to put in front of the CEO

A one-page policy should name the owner (marketing, with security and legal consulted), list each bot and its status, state the reason in one sentence, and set a quarterly review. Bring it to the CEO and finance as part of the CMO, CEO and CFO alignment conversation, since the choice affects both content risk and pipeline.

Do AI Overviews reduce website traffic

The two sides measure different things, so a CMO should read them as two partial views rather than one verdict.

In our reading, the two claims can both hold: more sites can get some traffic while any one site gets less.

How to read your own referral data

A worked example, not a benchmark: say you get 50,000 Google sessions a month and they fall 8 percent, which is 4,000 fewer sessions. If 20 of your 50 top pages show an AI Overview, 40 percent of that set is exposed. Check whether the loss concentrates there. If it does, we think AI Overviews are a likely factor. If it spreads evenly, look at seasonality, rankings or demand first, and pair the analysis with share of search to see whether demand itself moved.

The eight questions to ask your team, with good and weak answers

Copy the table below into your next marketing review and ask each question as written. Pay attention to how specific each answer is and how quickly it comes. A team that has done the work names pages, dates and owners. A team that has not talks in generalities.

How to measure AI search visibility and connect it to pipeline

The four metrics to report monthly

Freeze the prompt set for at least a quarter. If the team rewrites prompts each month, the trend line measures the rewrite.

A worked example, not a benchmark: take a fixed set of 50 buyer prompts. If you appear in 20 of the 50 answers, share of answers is 40%. If 8 of the 50 answers cite your pages, citation rate is 16%. The gap between those two figures is the work list for content.

The traffic line is growing fast enough to matter.

Connecting the signals to pipeline

No single report closes the loop.

A practical sequence:

1. Feed AI referral sessions and branded search into your marketing mix model as inputs, not outcomes.
2. Run a geographic or time-based incrementality test when you make a large content change, and compare branded search and marketing-sourced pipeline against a holdout.
3. Report the result as a range, with the gap stated. Purchases that start in an AI answer and finish on a marketplace will not show up in your own data.

Put all four metrics on one slide of the board deck, with the model's range beside them.

A worked example and the decisions only the CMO can take

Here is our method applied to an example business-to-business company. The inputs below are round example figures, not benchmarks.

The worked numbers

The team builds a fixed set of 50 buyer prompts and runs it every month across the AI assistants its buyers use. In the first run, the company's pages are cited in 8 of 50 answers, a cited-answer share of 16%.

The team then rewrites the pages behind the weakest prompts. Each page gets a plain definition near the top, a comparison table and named, dated proof. Three months later the same set returns citations in 20 of 50 answers, or 40%.

That is 12 more cited answers (20 minus 8) on an unchanged set. The prompt set has to stay fixed. If the prompts change, the trend line tells you nothing.

On the revenue side, the site records 50 AI referral sessions in a month, and 8 of them become qualified leads, a 16% session-to-lead rate. We would report that rate next to marketing-sourced pipeline. We would treat it as directional until an incrementality test supports it.

The three decisions only the CMO can take

What this page does not cover

This page does not compare monitoring tools or vendors, cover AI assistants beyond Google, OpenAI and Perplexity, or give traffic benchmarks for your category. It is not legal advice on crawler policy or content use, so involve your legal team when setting the policy.

Related on CMO Tribune: Marketing mix modeling, explained for the CMO who has to choose a measurement model; In-house vs agency cost, compared with your own numbers; Marketing team structure by revenue band: functional, pods, hub-and-spoke and business-unit models.