What does ChatGPT say when your buyer asks?
I ask ChatGPT, Gemini, Perplexity and Claude the questions your buyers ask, record every answer on the day I ask it, and hand you a dated report: where you are cited, where a competitor is recommended instead, and the three changes to make first.
$1,500 for the audit. Delivered in 48 to 72 hours. Nothing estimated, nothing scored by a black box: you get the answers themselves.
- Questions
- 40 or more, written in your buyers' words
- Assistants
- ChatGPT · Gemini · Perplexity · Claude
- Recorded
- verbatim, dated, one fresh session per question
- Read by
- a person, every answer
Four steps, all of them visible in the report.
The value is not a score. It is a record you can hand to your board, your agency or your SEO lead, with every claim traceable to an answer they can read.
Questions in your buyers' words
Forty or more questions across the stages a buyer goes through: discovery, comparison, evaluation, pricing, trust, shortlist. I write them from your category, your market and the competitors you name. You see the full list before I run anything.
One fresh session per question
Each question goes into the public app of each assistant in a new chat, with no memory and no custom instructions, from a clean account. The answer is saved verbatim with the date and the assistant. Answers vary by day and place; that is why every one is dated.
Read, not scraped
Software finds your brand and the tracked competitors in each answer and notes the position. I read every answer myself: how you were described, what the assistant got wrong, which source it seems to lean on.
A report you can act on
Share of voice by assistant, the questions where you are absent, who is recommended instead, and three changes ranked by effort and effect, each tied to a page or an entity on your site. Plus the raw answers, so nothing has to be taken on trust.
What you are paying for.
| 1 | Summary | One page. Where you stand across the four assistants, in numbers and in two sentences. |
| 2 | By assistant | Answers recorded, times you were cited, share of voice, competitors named. One row per assistant. |
| 3 | By question | Every question, every assistant: cited or absent, position, who else was named. This is the table people photograph. |
| 4 | Where you are absent | The questions that matter most in which you do not appear at all, and who fills the space. |
| 5 | How you are described | What the assistants believe about you: pricing, audience, strengths, errors. Verbatim quotes. |
| 6 | Three changes first | Ranked by effort and effect. Page-level briefs: which page answers which question, what the first 150 words should say, which entities and comparisons are missing. |
| 7 | Appendix | All recorded answers with dates. Reproducible: ask the same question the same way and you will see what I saw. |
Not in the report: a guarantee. Nobody controls what these assistants answer, and anyone who promises a position is guessing. What I can promise is that every line in the report is something I actually recorded.
Built for real brands, from public information only.
Each sample is a shortened audit prepared without the company's involvement so you can judge the actual output, not a showreel. They are labelled as samples, and any company can have its sample removed within a day.
Sample reports are being prepared. The first two are on Buffer (US) and Bloom & Wild (UK).
Three ways to buy it.
| Service | What is included | Price |
|---|---|---|
| Audit | 40 or more questions, four assistants, all answers recorded and read, the full report above, a 45-minute walkthrough call, the raw data. | $1,500one-time · 48-72 hours |
| Audit and monitoring | The same questions re-run every month. Movement by assistant, new competitors, new absences, and a fresh list of the three changes to make next. | $750 per monthafter the audit · cancel any month |
| For agencies | Up to five client brands, a shared question bank per vertical, reports under your name, one monthly review with you. | from $2,500 per monthwhite-label |
Half on order, half on delivery. Card or bank transfer in USD, EUR or GBP, invoice with every payment. If the delivered report does not contain the scope we agreed, I complete it at no charge or refund the deposit.
Dashboards need a reader.
Profound, Peec, Scrunch and the rest sell monitoring software. It is good software, and if you have someone whose job is to read it every week, use it.
Most marketing teams do not. They need, once, a clear answer to "where do we stand and what do we change", written by a person who read the answers, in a form the CMO can forward. That is what Cited is. If the audit shows you need weekly tooling afterwards, I will say so and name the tool.
Gabriel Homoky. I run Automating, a small studio in Europe that builds AI systems for businesses. Cited is the studio's research practice: one person, one method, one kind of report. I work over email with clients in the US, UK and Australia and answer within a business day.
I started Cited because every "AI visibility" conversation I sat in ended with screenshots of one ChatGPT answer and a lot of opinion. The fix was boring and obvious: ask many questions, record everything, date it, read it. So that is the product.
Straight answers.
Can you get us recommended by ChatGPT?
No, and neither can anyone else with certainty. What the audit does is show where you are missing, why the assistants seem to pick the competitor instead, and which of your pages should carry the answer. Those changes move results in practice, but the report never promises a position and the terms say so.
How is this different from running the prompts ourselves?
You can, and the appendix lets you check my work. The difference is volume and discipline: forty-plus questions across four assistants in clean sessions, recorded the same way, read by someone who does this every week and knows what a source pattern looks like. Most teams try it for an afternoon, get twelve screenshots and no conclusion.
Is the analysis done by AI?
The assistants are the subject of the study, and software counts mentions and positions. Every answer is read by me and every recommendation is written by me. The report says "AI-assisted, human-reviewed" and means it.
What do you need from us?
Your category in your own words, your markets, three to five competitors, and your website. No access to anything. I confirm the question list with you before running it.
How do you handle our data?
Everything used is public. The report is yours to share inside your company and with your agencies. I keep the recorded answers for the monitoring baseline and delete them on request.
Why 48 to 72 hours?
Recording 160-plus answers by hand in fresh sessions takes a day. Reading them and writing the changes takes another. Anything faster would be automation, and automation is what you are trying to get a second opinion on.