Your client’s page gets cited in ChatGPT. A competitor gets recommended. If the weekly report calls that a win without mentioning the shortlist, it misses the outcome the buyer came for.
Start with the distinction everything else follows from: a mention names the brand, a citation uses its page as a source, and a recommendation presents it as a fit for the buyer. These are three different outcomes, not three ways to say visibility. I’ll use one running example: a revenue attribution tool, ClientCo, trying to reach a 50-person team on HubSpot. Its educational page can answer a question beautifully and still fail to help when that team asks which tool to buy.
A citation can sit beside someone else’s shortlist
Take the prompt ClientCo cares about: best revenue attribution tool for a 50-person team on HubSpot? This is a supplier-selection question. The buyer needs a fit for a particular team and setup. ClientCo’s page explaining what revenue attribution means was written for a different job.
Hypothetical answer, invented to show the split: “For definitions of revenue attribution, see ClientCo’s explainer [citation to ClientCo]. For a 50-person HubSpot team that needs multi-touch reporting without a full data team, shortlist Competitor A and Competitor B, both listed in three independent best-of roundups and discussed in recent user threads.”
ClientCo supplied the definition. Someone else got the shortlist. Both outcomes sit in one response.
That split is not a reporting technicality. A Semrush and Kevin Indig study of 3,981 domain appearances found 61.7% were cited but never named, 25.1% named without citation, and only 13.2% both cited and named. Even the overlap between citation and name is limited. Neither outcome, on its own, tells you whether the answer recommended the supplier.
The first question for an agency is therefore not, “Did we appear?” It is, “What did the answer use us for?” In ClientCo’s case, the citation establishes that the explainer helped answer the background question. It does not establish that ClientCo made the buyer’s shortlist. Grade those separately.
Three outcomes, three different claims in the report
A source citation means the answer used your client’s page as evidence. It may footnote an explainer, quote a definition or link a how-to. I used to treat this as the hard part. I was wrong. On informational queries, the citation rate runs at 89.3% against an 18% mention rate. The answer’s job there is to explain, not to choose. ClientCo can write the clearest definition of revenue attribution and win the source slot without entering a buyer’s consideration set.
A brand mention means the answer says the client’s name. No link is required, and no endorsement is implied. Engines differ here. In that same dataset, Gemini named the brand in 83.7% of cases but linked in only 21.4%, while ChatGPT linked in 87% and named in 20.7%. Across prompts answered by several engines, they disagreed on naming in 22% of cases. A blended visibility score hides that difference. If ClientCo gets named in one engine but not another, the report should show both results, not average away the question.
A recommendation means the answer presents the business as a fit for that buyer. It needs a reason tied to the request: why ClientCo suits this HubSpot team, not merely what revenue attribution means. This verdict can appear without a link to ClientCo’s site. In a Position Digital study of 278 B2B prompts, ChatGPT named brands 552 times but cited brands’ own sites only 375 times; 66% of recommendations arrived without an own-site citation. Presence in best-of lists correlated strongest (r=0.64), followed by brand mentions and Reddit presence. An agency that checks only its client’s cited pages can miss recommendations as well as overstate them.

Why can one answer cite ClientCo and recommend another supplier? Because explaining a problem and establishing supplier fit are different jobs. The definition page helps with the first. To answer the second, the model needs grounds to judge category, constraints and corroboration. Match the metric to the buyer question before you call the appearance a win.
An explainer cannot prove what the offer page leaves out
ClientCo can define multi-touch attribution in 400 clean words and still leave a buyer guessing about who the tool serves, what it costs to run, what it will not do or what independent buyers say about it. None of that makes the definition bad. It makes the page insufficient for best tool for this team.
In the hypothetical answer, the competing suppliers have details the buyer can evaluate: a HubSpot native connector, onboarding in under two weeks, an exclusion for enterprise data warehouse builds, best-of list inclusions and specific user discussions. Those details do not turn a mention into an automatic recommendation. They give the answer a way to test fit that a definition alone cannot provide. For ClientCo, the useful brief asks for its accurate offer, markets, exclusions and supportable proof, not a longer opening paragraph on attribution.
Access comes before either kind of answer. If search bots cannot fetch ClientCo’s page, the page may not be available for ChatGPT, Claude or Perplexity to cite. That is why I check whether OAI-SearchBot, Claude-SearchBot and PerplexityBot are allowed in robots.txt while training bots such as GPTBot are handled separately. It is a quiet failure on a white-label account. But fixing access makes a page available as a source; it does not prove the company belongs on a shortlist. I keep our white-label glossary open when briefing writers for precisely that reason.
Free tool: the AI Crawler Checker shows which AI crawlers your robots.txt lets in and what they receive when they arrive.
The practical sequence is simple: make the right page accessible, give it a question it can answer, then check the outcome that question calls for. Technical access is a prerequisite, not a recommendation strategy.
Brief the educational question and the buying question separately
For ClientCo, I would write two briefs rather than ask one page to cover the whole journey:
- Educational brief: What is revenue attribution? How does multi-touch differ from first-touch? What breaks in HubSpot reporting at 50 seats? Build clear answers that can serve as sources.
- Supplier-selection brief: What is the best attribution tool for HubSpot? How does ClientCo compare with Competitor A? Who can onboard without a data team? Give the answer accurate grounds to assess ClientCo as a fit.
The first brief targets citations. The second targets recommendations. Neither guarantees its target. The point is to stop assuming that a 2,000-word guide which explains the category also proves the supplier. When those jobs get blended, an agency can produce a useful source and still leave the buying question unanswered.
The supplier brief is shorter and harder. It needs the real plans and onboarding time, connector and seat minimums, service area and support hours, and explicit exclusions such as custom warehouse builds. If any of those details do not apply to ClientCo, leave them out. Then identify where outsiders can confirm the relevant claims through best-of lists and specific user threads. An attractive comparison table cannot substitute for proof that does not exist.
This is where white-label work becomes a program rather than a page order: entity work, extractable definitions and comparison tables, llms.txt and schema, trusted-source citations, and prompt-level tracking. AEO work aimed at citations and GEO work aimed at visibility and recommendation should not collapse into one deliverable called AI content. Assign the gap to its owner. Content addresses the explainer. Technical work addresses access. Corroboration addresses the missing grounds for a shortlist. Ordering another undifferentiated page addresses none of them.
Make the weekly report answer the buying question
A single screenshot is a poor measurement plan. Freeze a panel of real purchase questions, comparisons, situations and fact checks, then repeat each prompt at least 10 times per engine. Log whether the answer names the brand and whether it links the client’s website separately. If it recommends ClientCo but cites a directory roundup, record a recommendation and a third-party citation, not an own-site citation. That distinction is part of the weekly-report method and prevents a directory appearance from being credited to ClientCo’s content.
Put the count beside the rate. “6 of 10” tells the reader more than “60%” alone: the 95% Wilson interval on 6 of 10 is roughly 31 to 83%. One run is a record of an answer, not a reliable verdict on the program. I run the gap audit first to identify commercial questions where competitors get cited instead of us. Then I watch whether repeated runs change the recommendation rate on those same prompts. Question type matters here too: comparative best and versus queries earn mentions at 2.4 times the rate of informational ones. Do not judge a definition page solely by a supplier-selection metric, or mistake its citations for supplier selection.
Keep the engine, exact prompt and observation date visible. ChatGPT and Gemini do not behave identically, so one blended score is not enough. Where measurement permits, put enquiries next to the visibility trend. If ClientCo moves from named in 2 of 10 to 6 of 10 on the HubSpot shortlist prompt but demo requests from organic and AI-assisted sessions do not move, say that. The method for measuring recommendations across engines with fixed buyer prompts is useful because it leaves room for that honest result.
The report now has a purpose: it shows whether ClientCo is being used to explain attribution, being named as a supplier or being presented as a fit. Keep all three lines, even when one makes the others look less impressive.
Apply the model to the next account
Suppose ClientCo appears in 8 of 10 educational answers as a cited source but in only 1 of 10 shortlist answers as a recommendation. I would not order more explainers. I would bring a narrower request to the strategist: pause that queue, make the offer page answer the HubSpot buyer’s constraints, and spend the next two weeks pursuing corroboration that can be checked elsewhere. That means an accurate scope block covering the applicable plans, onboarding time, connector, seat range and exclusions, alongside the proposed two independent roundup or thread placements that name ClientCo.
I run this inside groas because the work happens daily, actions are logged and a named strategist owns the result. That is a better operating model than a monthly white-label PDF that counts the citation and stops there. The SEO and AI Search service runs for one flat fee and keeps paid and organic context together when the buyer question touches both ads and attribution.
Now change the category and see whether the distinction still holds. Say the next white-label account is a plumbing company in Austin. Its blog gets cited for why does water pressure drop in summer because it explains pressure regulators clearly. A buyer then asks: who can replace a 50-gallon water heater in Austin this week? The answer cites the pressure article as background and recommends two competitors that list same-day installation, service area by ZIP, price range, licence number and 200 recent reviews with photos.
Run the three tests. The pressure article is cited because it helps explain a plumbing issue. The company may be mentioned if the answer says its name; the citation alone does not tell us. It is not recommended for the water-heater job in this answer. The buying question asks about a different service and a deadline, so a clear explanation of pressure regulators cannot establish fit. The analogy to ClientCo holds at that point and stops there: the plumber’s service area and availability are not a software tool’s connector and seat range. The questions differ; the reporting distinction does not.
Keep the explainer. Brief a separate water-heater selection page with the company’s accurate offer, markets, exclusions and supportable proof. Pursue the proposed independent mentions that can confirm it, then track citations, mentions and recommendations across 10 repeated runs. Do not claim the new page won the shortlist until the answers show it.
For the agency owner opening the Monday report, the check is three lines:
- Cited: In how many of 10 runs did the answer use the client’s page as a source?
- Mentioned: In how many of 10 runs did it say the client’s name?
- Recommended: In how many of 10 runs did it present the client as a fit for that buyer?
Attach the engine, prompt and date to each line. When the third lags the first two, do not order more explainers. Fix scope, fix proof and pursue corroboration elsewhere. That is the difference between reporting visibility and reporting a path to revenue, and it is the standard I hold groas white-label work to: every action logged, with one strategist accountable for movement in those three lines.

