Run the same fixed set of buyer questions across ChatGPT, Claude and Perplexity on a repeating schedule and count your name's appearance each time; one check only tells you today's answer. AI models rarely return the identical brand list twice on the exact same prompt, so a single run is a sample, not a measurement (SparkToro, 2026).
It means fixing a set of real buyer questions and asking them to the same engines on a schedule, then counting how many times your name comes back each round. The unit you count is a Buyer Question answered by one engine on one run, not a single blended "visibility score." Our comparison of the leading AI visibility tools covers ten vendors that help run this at scale, but the method works the same way by hand: five to ten real buyer questions, asked to ChatGPT, Claude and Perplexity, repeated on the same day each week.
Bottom line: the thing worth tracking is a count per engine per run, logged over time, not a single number read once.
Three things move the count on their own: the model, the sources it is drawing on, and your competitors' content. AI labs update models on their own schedule, which can shift which sources an engine treats as authoritative. Engines with live web search re-crawl pages continuously, so a competitor's new comparison page can enter the answer pool within days. And repeated testing of language models shows answer-consistency between identical prompts is routinely short of 100%, so some of the weekly wobble is simply how the system works.
Proof: arXiv — The Non-Determinism of LLMs: Evidence of Low Answer Consistency in Repetition Trials.
Why it matters: most of what moves your count is outside your control and invisible until you check again, which is the actual argument for a schedule instead of a single read.
Treat one or two weeks moving as noise, and wait for the same direction across at least three to four consecutive runs before you call it a trend. A single week down is normal; a count that keeps sliding for a month, on the same question set and engines, is the signal worth acting on. This is the same discipline that makes a share-of-voice number trustworthy: fix the questions, fix the engines, compare across time only once both are held constant.
Do this: log every run in one sheet or dashboard, even the weeks nothing seems to have changed; the flat weeks are what make a real move visible.
One bad week on one engine is noise; the same competitor replacing you on the same question across two or three engines, for two or more weeks running, is a real signal. In one Reflexa evidence-log example, a CRM buyer question kept naming HubSpot first across ChatGPT, Claude and Perplexity in every run, while Salesforce, Pipedrive and Zoho rotated through the second and third spots depending on the engine and the week (Reflexa AI Visibility Index, October 2026). Read the rotation as normal; read a competitor consistently moving into first place as the thing to investigate.
Proof: Reflexa AI Visibility Index — CRM: who AI recommends, October 2026, evidence log.
Rule of thumb: one engine, one week, one question moving is not a trend; the same loss repeated across engines and weeks is.
Most AI visibility tools can run a single check; fewer keep the same question set on a repeating schedule and let you see the count move run over run. Here is what each vendor's own pricing or documentation page says, read in September 2026 unless noted:
| Tool | Engines measured | Repeat / weekly tracking | Starting price |
|---|---|---|---|
| Reflexa | ChatGPT, Claude, Perplexity | Yes, weekly, on every paid plan; full answer + sources kept per run | $49/month, first check free |
| Profound | ChatGPT, Gemini, Google AI Overviews on trial; Claude on paid tiers | Advertised on paid tiers (tryprofound.com/pricing) | see vendor pricing page |
| Otterly.AI | Core engines on entry tier; Claude as a paid add-on | Advertised on paid tiers (otterly.ai/pricing) | see vendor pricing page |
| Semrush AI Toolkit | ChatGPT, Google AI, Gemini, Perplexity; no Claude at any tier (semrush.com/pricing/ai) | Advertised as part of the toolkit | see vendor pricing page |
| Peec AI | Not confirmed on the vendor's own page this run | Not confirmed on the vendor's own page this run | see vendor pricing page |
| Ahrefs Brand Radar | Not confirmed on the vendor's own page this run | Not confirmed on the vendor's own page this run | see vendor pricing page |
Note: "offers tracking" and "attaches the actual answer to each run" are different things; see which tools show the real answer, not just a score before you pick one to track with.
You can start tracking your own count without buying anything yet:
Reflexa's agent runs a page check on your site every week and keeps every change on record, so when a count moves you can see whether the site moved first.
Fix the same buyer questions and the same engines every time, run them on a schedule rather than ad hoc, and look at several consecutive runs before calling a move a trend rather than noise, since a single run is only one sample from a non-deterministic system.
AI labs update their models on their own schedule, engines with live search re-crawl pages continuously so a competitor's new page can enter the answer pool within days, and the models themselves rarely return an identical answer to the same prompt twice.
Treat one or two runs as noise and look for the same direction across at least three to four consecutive weekly runs before you call it a real shift; a single week moving is normal, a sustained multi-week slide is the signal worth acting on.
Yes. The first check at dashboard.reflexatechnologies.com/audit is free, takes minutes and needs no card; it gives you one dated reading you can repeat by hand weekly, or move to a paid plan once you want the schedule run automatically.
The free check reads your pages the way ChatGPT, Claude and Perplexity read them — 23 structural checks, evidence attached, no card needed.