Ask ChatGPT, Gemini, Claude, Perplexity, Grok, Le Chat and DeepSeek the questions your buyers actually ask, and see who they name. citecheck runs 30 buyer-intent questions across seven assistants, scores how often you're recommended rather than merely listed, and shows which competitors got the recommendation instead.
The query set changes shape with what you are. One from each, to show the difference.
Switching, pricing, what the incumbent gets wrong.
“best screenshot tool for mac without a monthly subscription”
Trust, rates and fit — people hiring, not buying.
“how do I find a good immigration solicitor and what should it cost”
Local, and driven by the occasion.
“anywhere good for breakfast in Shoreditch that isn't a standard fry up”
A worked example — Shottr, a Mac screenshot app. Tap any cell to read the answer behind it.
“best screenshot tool for indie hackers”
You were not mentioned in this answer.
What the assistants read before answering about your category. Not advice — a list. Get onto the ones you are missing and the next run moves.
Illustrative. Your report lists your own, counted across all 210 answers.
Re-run weekly. Hover a point for that week.
▼ 6 pts in 8 weeks
▼ 5 pts in 8 weeks
▲ 20 pts in 8 weeks
▲ 19 pts in 8 weeks
▲ 5 pts in 8 weeks
▼ 6 pts in 8 weeks
no trend — not scored
DeepSeek sits apart: its API cannot search the live web, so it is shown but not scored.
Perplexity recommends you more often than anyone.
For solo developers, CleanShot X and Xnapper are popular, but Shottr stands out for its one-time pricing and scrolling capture.
Your product name never appears in the ones that test discovery. Edit any of them first.
Each question, every engine, web search on wherever the API offers it.
Recommended first, recommended, listed, cited, or absent — with the sentence that decided it.
Being listed among ten tools is not the same as being the answer.
Worth 3, 2, 1, 0.5 and 0 points respectively.
Two of these were never on the customer's competitor list.
Six kinds. Three may never contain your product name — naming you would only test whether the assistant can read.
“what is the best lightweight screenshot tool for mac that doesn't have a monthly subscription”
“my mac's screenshot tool is too slow for design work, is there something tiny and fast”
“CleanShot X vs Xnapper, which is better for frontend developers”
“what is a good alternative to CleanShot X that isn't a subscription”
“how do I take a scrolling screenshot of a long webpage on my mac”
“is Shottr free or paid and where do I download it”
Chosen by the pattern we actually found. Each one phrased as a test.
Front-load direct answers. Put hard numbers where they can be quoted.
Re-run in 4 weeks · cited vs recommended
OpenAI leans on Bing. Submit your sitemap to Bing Webmaster Tools and confirm you are indexed.
Re-run in 4 weeks · ChatGPT score alone
Nothing to quote when someone asks X vs Y. Publish an honest comparison page that concedes what you are worse at.
Re-run in 4 weeks · comparison queries
Engines select differently. Treat the weak one as its own problem, not an average to raise.
Re-run in 4 weeks · per-engine spread
Profound, Peec and Scrunch do this for brand teams, at brand-team prices. Start free, in two minutes, without talking to anyone.
$0
$19/mo
$49/mo
$149
The free tier covers ChatGPT, Gemini and Perplexity. Paid tiers add Claude, Grok, Le Chat and DeepSeek.
Ask it the questions your buyers ask — not your product name — and see whether you come up. citecheck does this across 30 buyer-intent questions and four assistants, then scores how often you're recommended, merely listed, or absent.
No. There is no ranked list of ten blue links to place in. An assistant returns one answer naming two or three products. You are either in that answer or you are not, so the unit of measurement is the recommendation, not the position.
It uses each vendor's API with web search enabled, which approximates the consumer app but does not match it exactly. Every result is labelled with the engine and that caveat. We would rather be accurate about the method than overclaim.
Every report ends with recommendations tied to the specific failure pattern found, each phrased as a test: do this, re-run in four weeks, compare. No tactics we cannot measure.
The report says the call failed. A failed call is never scored as absent — that would tell you the AI doesn't know you when in fact nobody asked it.
Yes. The question shapes differ — a place is asked about locally and by occasion, a service by trust and price — so the generated query set differs too. Anything people ask an AI to choose between can be audited.
Every answer an assistant gives cites what it read. We aggregate those across all 210 answers, so the list is what actually informed the recommendations, not a guess at what might matter.
The first audit is free and needs no account. Tracking starts at $19 a month for weekly re-runs across all four engines.
The first audit is free and needs no account. You'll see your score, the questions you lose, and who wins them instead.
Run a free audit