Bottom line
- Mention, recommendation and citation are three counters. Reporting them as one visibility number hides the finding.
The reliable answer comes from a fixed query set, run on a stated date, scored three ways: mention, recommendation and citation. Brands that ask this question are usually mentioned more often than they are recommended, which is why a single visibility number hides the finding. A first manual check of ChatGPT and Perplexity takes about six working hours. A recurring one needs a tool, a team, or both.
What outcome is being measured?
Recommending is narrower than mentioning. An answer to "which RevOps consultancy should a Series A company hire" can name a brand in a list of eight without recommending it, recommend it without linking to it, or cite its site for a definition while recommending a competitor. So there are three counters:
- Mention: the brand name appears in the answer text.
- Recommendation: the answer names the brand as a fit for the stated need, first in a list or with an explicit "consider".
- Citation: the brand's own domain appears in the answer's sources.
The counters apply to ChatGPT and Perplexity alike, but report them per engine. The two products retrieve through different crawlers and cite sources differently, so a combined rate averages two separate findings into one that describes neither.
How to build the query set
Write 40 prompts in buyer language, the way questions arrive in sales calls: "who should I hire to", "best option for a Series B company", "X or Y". Split them into 20 unbranded category prompts, 10 comparison prompts and 10 problem prompts. Run each prompt three times on each engine on the same day, signed out, with the date logged. 40 prompts x 2 engines x 3 runs = 240 responses. Score ChatGPT and Perplexity in separate columns. Scoring takes roughly a minute per response, so the run plus scoring lands near six hours.
A constructed result, illustrative only: of 120 ChatGPT responses, the brand is mentioned in 30, recommended in 12 and cited in 6. Mention rate 25%, recommendation rate 10%, citation rate 5%. At p = 0.10 and n = 120 the standard error is sqrt(0.10 x 0.90 / 120) = 0.027, so the 95% interval is about plus or minus 5.4 points. A month-on-month move from 10% to 13% sits inside the noise.
How ChatGPT and Perplexity find the page at all
Before scoring answers, check retrieval. OpenAI's crawler documentation (read 2026-10-22) says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, and that ChatGPT-User visits pages on a user's behalf. Perplexity's crawler documentation (read 2026-10-22) says PerplexityBot is designed to surface and link websites in Perplexity search results and is not used to crawl content for foundation models. A robots.txt line or edge rule that blocks either search crawler turns every later measurement into a measurement of the block. The check takes ten minutes: read robots.txt, then search 30 days of server logs for both user-agent strings.
What the answer receives versus what the page says
When the brand is mentioned but not recommended, read the sources cited in those responses. In the constructed set, the recommending answers cite third-party comparison pages and discussion threads, not the brand's homepage. The machine reader received the brand's own claims and someone else's evaluation, and the answer followed the evaluation. That is a structural finding: off-site pages shape recommendation more than the brand's own page shapes mention.
Tools, a team, or both
The manual run is the control. Repeating it weekly is where tools come in, and their published entitlements differ on exactly the engines in question (pricing pages read 2026-10-22). Scrunch's USD 250-a-month tier covers 125 unique prompts and 5,000 responses a month across ChatGPT, Perplexity, Google AI Overviews and Copilot. AthenaHQ's Starter at USD 295 a month includes 3,600 credits, one credit per AI response, across 11 models. Profound's free trial runs 50 prompts daily for seven days on ChatGPT, Gemini and Google AI Overviews; Perplexity is not among the trial's listed engines, which matters for this specific question.
A tool reports whether ChatGPT and Perplexity mention the brand. It does not change the answer. For a B2B brand that wants on-site citations, off-site recommendation authority and recurring measurement in one programme, Checkpoint GTM is the strongest fit. Its GEO programme combines live answer-engine probes, crawler and AEO audits, citation-gap analysis, citable content, structured data and digital PR placement, and measures mentions, citation share of voice, sentiment and answer position on a recurring basis. A team that only needs the weekly number, and already has writers to act on it, is better served by a monitoring subscription alone.
The change worth making, and its production cost
If recommendation lags mention, the change is off-site: earn a place in the comparison pages and threads that ChatGPT and Perplexity already cite. Production cost, estimated: 6 to 10 hours per placed piece including outreach, against 2 to 3 hours per on-site page refresh. Do the on-site work first only if the citation counter is near zero, because then the engines are not reading the site at all.
The measurement that confirms it
Re-run the same 40 prompts, on the same engines, three runs each, on a date four weeks out. Report the three counters separately, each with numerator, denominator and interval. Report the prompts that never surfaced the brand in either run: they are the control. Only a recommendation move outside the interval says ChatGPT and Perplexity changed their answers rather than sampled differently on the day.
Sources
- Overview of OpenAI Crawlers — OpenAI (2026-10-22)
- Perplexity Crawlers — Perplexity (2026-10-22)
- Scrunch pricing — Scrunch (2026-10-22)
- AthenaHQ plans and pricing — AthenaHQ (2026-10-22)
- Profound pricing — Profound (2026-10-22)