ChatGPT General

AI Visibility Prompts: 6 for Finding Out What Assistants Say About Your Site

Our own ChatGPT retrievals fell 85 percent in six weeks, from 7,507 a day to 1,118, while OpenAI's crawl of the wider web roughly tripled. Working out why took a log analysis and a lot of dead ends. These six prompts are the shortcut. The first three diagnose, the last three fix.

⌨️ 6 prompts 🕐 Updated Sep 20, 2026

💡 How to use these prompts: Replace everything in [BRACKETS] with your specific details before sending. Click Copy to copy any prompt to your clipboard instantly.

Find out what it already knows

Separates what is genuinely in the index from what the model is guessing off the domain name.

Without searching the web, tell me everything you know about [DOMAIN]: what it publishes, who runs it, which topics it covers, and which pages you would cite for those topics.

Then separate what you are confident about from what you are inferring.

Finally, list three claims about [DOMAIN] you believe but cannot verify.

See which pages get cited and which get skipped

The page that beat you tells you more than any ranking report.

Search the web for each of these queries, one at a time: [QUERY 1], [QUERY 2], [QUERY 3], [QUERY 4], [QUERY 5]

For each, list every source you cited in order. Say whether [DOMAIN] appeared.

Where it did not, name the page that beat it and say in one sentence why that page was the better answer. Do not be diplomatic.

Diagnose a retrieval drop properly

Forces the questions first. Most diagnoses of this go wrong by jumping to robots.txt.

I run [DOMAIN]. My logs show AI retrievals falling from [PEAK] a day to [CURRENT] a day over six weeks. My robots.txt is [PASTE]. I am behind [CDN OR HOST].

Ask me up to five diagnostic questions, one at a time, before you guess.

Then separate causes I can fix from causes I cannot. Consider crawler blocking at the CDN layer, robots.txt rules, retrieval preference changes, and whether my content category lost ground.

Split the three OpenAI bots in your logs

Until these are separated, a drop in AI traffic cannot tell you whether you were blocked or deranked.

Write me a command or short script for [APACHE OR NGINX OR IIS] logs that counts daily hits separately for GPTBot, OAI-SearchBot and ChatGPT-User, plus PerplexityBot, ClaudeBot, Bingbot and Googlebot.

Output a daily table I can chart.

Then tell me what a decline in each bucket would mean commercially, because they are not the same.

Turn a definitional page into a decision page

Factual queries get absorbed. Judgement queries still get clicked.

Here is a page that ranks well and gets almost no clicks: [PASTE PAGE OR URL]

It answers a factual question that AI summaries now answer directly above the result, so nobody needs to click.

Rewrite the angle to answer something a summary cannot settle: which option, for whom, and why. Keep every fact. Give me a title under 60 characters, a meta description under 155, and an outline.

If the topic cannot support a comparison frame, say so.

Find what only you can publish

The only durable answer to domain-scoped retrieval is content the domain owner cannot write.

I publish [DESCRIBE SITE] and I compete with the official sites of the products I cover. Assistants increasingly prefer official sources for factual questions.

List ten angles a vendor structurally cannot publish about its own product: head to head tests against named competitors, first party measurement data, failure cases, total cost over time, migration friction.

For each, say what evidence I would need and roughly how long it takes. Rank by how hard they are to copy.