AI Search Reporting for CMOs: Mentions, Citations, Recommendations and Leads
Eric Strate / Insights & Research
A visibility percentage can be useful, but it cannot tell a CMO whether people are choosing the business, whether a website is being cited, or whether any opportunity enters the pipeline. This guide defines the different outcomes so agencies, software vendors, and marketing teams can report the same thing in comparable terms.
Four outcomes that should never be treated as one
| Outcome | What counts | What it does not establish |
|---|---|---|
| Brand mention | An AI response names the company or a reliable alias. | A source citation, positive endorsement, click or sale. |
| Source citation | The response explicitly links or attributes a cited URL or document. | That the brand was recommended, or that the content caused a conversion. |
| Buyer recommendation | The response presents the company as an option that fits the user’s stated need. | A guarantee that a user saw, trusted or selected it. |
| Qualified inquiry | A lead meets a pre-agreed business qualification rule and can be connected to available interaction evidence. | Automatic proof that one earlier citation caused the inquiry. |
A company can be cited as an information source without being suggested as a vendor. Another provider can be recommended without a direct link to its website. A buyer can also learn a brand name in an AI answer and arrive later as Direct traffic. These paths are not interchangeable.
Define the denominator before presenting a percentage
For a controlled prompt panel, a useful basic measure is responses with a brand mention divided by the total completed responses in the panel. Report the total prompts, repetitions, models, time window and geography. If you test 20 different prompts on three systems once each, you have 60 sampled responses, not a representative survey of every consumer search.
Keep incomplete or unavailable conversations separate. Do not assume that a model that exposes no source trace performed no retrieval. Record exact cited URLs separately from mentions. When competitors share a response, brand totals may exceed the number of responses, so a leaderboard percentage should not be interpreted as conventional market share.
What a real pilot result looks like
in an October 28, 2026 Gumshoe agency-selection audit of Eric Strate, 60 sampled conversations included one response mentioning Eric Strate and 26 mentioning iPullRank. The dashboard also exposed one citation to ericstrate.com/wordpress-aeo/ for a question about an AEO/GEO agency handling structured data for a WordPress content site.
That is useful evidence that the page entered at least one answer’s citation set. It does not demonstrate a permanent ranking, client outcome, or a causal effect of the WARP markup that appears on the page. this is a first-party observation, not independently verified third-party performance. See the full citation pilot and limitations.
The weekly source ledger an agency should keep
For each tested answer, store the exact prompt text, prompt cohort, buyer role, requested geography, platform, date and time, whether retrieval was enabled, full visible answer, brand mentions, citations, recommendations, and any unavailable fields. A short source ledger should identify the cited domain and page, not merely the number of citations.
Record significant content changes separately with the page URL and deployment date. If an external agency publishes a case study, capture its public URL and evidence type. If an AI answer cites that case study, label it as an observed citation rather than proof that the external article caused the result.
A monthly CMO scorecard
| Question leadership asks | Report |
|---|---|
| Are we appearing for important buyer questions? | Branded and unbranded mentions by prompt group, platform and geography. |
| Are buyers getting our name as a viable option? | Explicit recommendations with context and dated example answers. |
| Are our pages being used as evidence? | Exact cited URLs, citation frequency and which question triggered each one. |
| Where do competitors win? | Competitor recommendations and the sources supporting them, with mismatched service models flagged. |
| What changed on our site? | Approved deployments, redirects, indexing changes and newly published evidence. |
| Is there business impact? | Identifiable AI referrals, assisted journeys, inquiries and lead quality, with attribution gaps explicitly noted. |
What can Google Search Console and analytics actually tell us?
Search Console can describe Google Search performance under its available reporting surfaces and dimensions. Analytics can identify some AI-assistant referrals when referrers or query parameters are preserved, and it can measure visits and events that actually reach the site. Neither automatically gives a complete list of all prompts that people asked private AI assistants or proves the origin of Direct visits.
Use those data sources as complementary evidence, not as replacements for answer-level research. Include the number of qualified contacts rather than optimizing only the number of citations. When attribution is uncertain, state the uncertainty.
How to compare platforms without misleading the client
Keep ChatGPT search responses, Claude web responses, Perplexity citations and Google AI Overviews distinguishable in the raw ledger. Their source access, presentation and sampling behavior differ. If a vendor compares results across services or months, ask whether the exact prompts, model settings and collection methods remained comparable.
Run an unchanged holdout cohort alongside newly selected prompts. Avoid editing the test questions merely to make your agency appear more often. Report both favorable and unfavorable examples, especially when they concern the most commercially valuable questions.
When implementation matters more than the dashboard
A dashboard describes observations. Search visibility work may require engineering, content, entity and linking changes by someone who can access the website. A managed provider should explain who owns those changes, how they are approved, and how the result is verified. the Eric Strate AI Search agency page describes one hands-on service model; the AI visibility diagnostic page discusses common reasons a business may be missing from answers.
Sources and methodological limits
Google’s AI features guidance describes the underlying eligibility rules; OpenAI’s ChatGPT search guide explains the search experience. For the research definitions used here, read RankGen’s published methodology. These sources do not establish that any agency can guarantee source selection.
