Free for marketers, founders and agencies
Does AI recommend your brand, or your competitor?
AI Visibility Desk is a free AI visibility tracker by Przemysław Zalewski, an independent engineer.
Every week it puts your buyers’ questions to Claude with web search on and shows whether your brand is named, who is named instead and which pages are cited (GEO).
What a report looks like
A real check from the test set: an invoicing app in Poland, two buyer prompts, three repetitions each. Every line is what the assistant returned, unedited; brand names are matched in the text, not judged by a model.
| Buyer prompt | Brand named | Competitors named | Pages cited every time |
|---|---|---|---|
| Jaki program do fakturowania z KSeF polecasz małej firmie? | 3/3 Fakturownia | wFirma (2/3), Faktura.pl (2/3), iFirma (1/3), inFakt (1/3) | taxeo.pl, niepoddawajsie.pl, eprzedsiebiorca.com, fakturbox.pl |
| Najlepszy program do wystawiania faktur online dla jednoosobowej działalności | 3/3 Fakturownia | inFakt (1/3), wFirma (3/3), Faktura.pl (3/3), iFirma (1/3) | pit.pl, mailgrow.pl, portal.faktura.pl, programydofakturowania.net, hauerpower.com, byblossom.pl, taxeo.pl, myslecjakmilionerzy.pl, sprytnafirma.pl, niepoddawajsie.pl |
How it works
Your brand and buyers
Brand name, website, up to 5 competitors and 10 questions a buyer would ask. The tool can suggest the questions from your website.
A check every week
Each question goes to the assistant with web search on, from your country. One answer per question, saved with the cited pages.
Trend, rivals, sources
Share of answers that name you, week by week; which competitors are named instead; which sites the assistant trusts, so you know where to get listed.
Measured before it was built
The test set is small and the engines change their answers between runs. That is the point: the tool repeats the questions every week and shows the share over time, with the full answers, so you can check every number yourself. One engine (Claude with web search) is live today; ChatGPT, Perplexity and Gemini are planned.
Is your site readable by AI?
The same app checks your website once: which AI crawlers your robots.txt allows (OpenAI, Anthropic, Perplexity, Google), whether you publish llms.txt and a sitemap, and whether your home page has a title, description, structured data and a canonical link. When a robots.txt rule looks like a block, it shows which paths it covers and whether any page from your sitemap falls under them. Each check links to the vendor documentation it comes from, and one button copies the findings as instructions for Claude Code, Cursor or another AI agent working in your site’s code.
Related: measure AI assistant traffic in PostHog with Dashboard Desk, and test your own AI prompts before release with Eval Desk.
What happens to your data
- Your brand, competitors and prompts are sent to the model (Claude by Anthropic, through Cloudflare) with web search on. Answers are stored for 180 days so you can see the trend.
- Weekly checks stop when you pause the project or delete the account. A hashed usage counter stays 62 days so the free limit survives account deletion.
- Nothing is shared with anyone. The weekly email can be turned off with one click.
Want to be the brand that gets recommended?
The free tool shows where you stand. Getting recommended is work: the pages assistants cite, content that answers buyer questions, structured data, and measurement in PostHog so you see which answers bring visitors and signups. I do that as a growth engineering engagement, starting with a written plan based on your report. I reply within 24 hours on working days, personally. If I am late, the first hour of work is free.
Questions
Which assistants are checked?
Today: Claude with web search on, which is how buyers use it. ChatGPT, Perplexity and Gemini are planned; each needs its own API account and will be added one by one.
Why does the result change from week to week?
The assistant searches the web and writes a new answer every time. Brands on the edge appear in some answers and not others. The tool shows the share over several weeks, so one odd week does not mislead you.
How is "named" decided?
Your brand name and aliases are matched in the answer text, as whole words, without a model judging it. You can open every answer and check.
Is it really free?
Yes: one brand, 10 prompts, each asked with and without web search, a check every week, 120 model calls a month. It is how I show what I build. Larger scopes and the work of getting recommended are a project.
How is it different from Otterly or Peec AI?
Otterly and Peec AI are paid trackers that cover several AI engines. This tool checks one engine today (Claude with web search), shows every full answer with the pages it cites, so you can verify each number, and is free for one brand and 10 prompts. Details in the Peec AI and Otterly comparison.
Can an AI agent fix what the check finds?
The technical part, yes: robots.txt, llms.txt, title, description and structured data. The app copies the findings as instructions that tell the agent what to change, what to leave alone, and to stop before writing content. Content written at scale by agents is what Google’s spam policies call scaled content abuse, and it hurts both search and AI answers, so ask me before you let an agent write pages.
Who built it?
Przemysław Zalewski, an independent engineer with 15+ years of shipping software and growth work measured in PostHog. A case study is on the PostHog page.
