How to Run an AI Visibility Audit by Hand

Run an AI visibility audit by hand when you want to know, in plain terms, which AI systems can already see your site and which ones are ignoring it.

This is a manual check. No software needed. You just need access to your server logs, a browser, and about an hour. The goal is to answer three questions: who is crawling you, what can they actually read, and are they sending you any traffic. You cannot improve what you do not measure, and this audit is the measurement.

Start with your server logs

Your server logs are the ground truth. Every crawler that visits your site leaves a record there. Open your raw access logs and filter for known AI crawler user agents.

Check for these specific agents first because they are the ones that matter most for visibility:

• ChatGPT-User, the live fetch agent for OpenAI, category ai_answer
• OAI-SearchBot, OpenAI's search index crawler, category indexing
• Claude-User, Anthropic's live fetcher, category ai_answer
• Claude-SearchBot (note: this is Anthropic's indexer, separate from Claude-User)
• PerplexityBot, Perplexity's indexer, category indexing
• Perplexity-User, Perplexity's live fetch agent, category ai_answer
• GPTBot and ClaudeBot, the training crawlers that send no traffic but still consume your bandwidth

You are looking for a few things in the logs. First, which agents appear at all. Second, which pages they requested. Third, how often they return. A single visit from ChatGPT-User means someone asked a question that triggered a live fetch of your site. Regular visits from OAI-SearchBot mean you are in the indexing pipeline.

If your logs show nothing from any of these agents, that is a result too. It means you are invisible to AI retrieval so far. That is not a judgment, it is your starting point.

Check your robots.txt against the real crawler list

Your robots.txt file is the gate. If you are blocking the wrong tokens, you are invisible for the wrong reasons.

There are three categories of AI crawlers, and they serve different purposes. CCBot (Common Crawl) is a training crawler that gathers bulk data. Blocking it removes your content from Common Crawl's training set. But blocking a training crawler does not remove your site from that operator's AI answers, because the indexing and live-fetch agents are separate tokens. If you block GPTBot but leave OAI-SearchBot open, OpenAI can still index you for search. If you block ChatGPT-User, you block live fetches made when a user asks a question in that moment.

The manual audit step is simple: pull up your robots.txt and compare it against the current list of crawlers. Ask yourself whether you meant to block each token. Many sites block training crawlers to save bandwidth, which is fine. Just know what it does and does not do.

Also check whether you have disallowed Google-Extended. That is not a crawler. Google-Extended (a robots.txt control token, not a crawler) controls how content Google already crawled may be used for AI training. It never visits your site. It has no user agent. It shows up in no logs. Disallowing it affects how Google uses existing data, nothing else.

One more thing to verify: Perplexity has no training crawler. PerplexityBot indexes and Perplexity-User fetches live. There is no third agent. If your robots.txt blocks PerplexityBot thinking it is a training bot, you are blocking the indexer that could bring you referral traffic. Perplexity does not build foundation models, so both of its agents are retrieval.

Test live fetches with a private browser window

Logs tell you what happened in the past. To know what an AI assistant sees right now, you need to ask some questions yourself.

Open a private window in a browser you do not normally use. Go to ChatGPT, Claude, and Perplexity. Ask each one a question that your business should plausibly answer. Use the phrasing a customer would actually use. Not "what is the best plumbing service in Portland" if your business is a niche B2B software tool. Ask something specific that your site genuinely addresses.

Then watch what the assistant does. Does it name your site directly? Does it cite a page of yours with a source link? Does it describe your product at all? Record the answers.

This is the part of the audit that tells you about real visibility. A crawler visiting your site is a signal of interest. A live fetch that results in your page being cited is the outcome you actually want. If you are not being cited in answers, you need to understand why the assistant chose another source instead.

If you want a deeper understanding of the mechanics behind this, our guide on how to optimize for AEO breaks down the 7 steps without needing to hire a consultant.

Trace the path from a query to your site

When you do get cited, you need to know the path. The manual method is clunky but workable.

Pick one answer where an AI assistant cited your site. Follow the source link. Check whether the page loads cleanly and quickly. Your site does not need Kymo installed for this. Any standard analytics will show you the referral traffic when an AI assistant links to you.

The realistic outcome is that you will see an uptick in referral traffic labeled from chat.openai.com, perplexity.ai, or claude.ai. Those are the confirmed click-throughs. Track how many there are per week and which pages they land on.

Here is the comparison that matters for your audit:

Your manual audit should classify every crawler you find into one of these three buckets. Training crawlers cost bandwidth and send nothing back. Indexing crawlers work ahead of time to build a search index. Live fetchers respond to a real person asking a question right now. The live fetchers are the most direct path to traffic, but the indexers are what make those live fetches possible.

Decide what "good" looks like for your size

A solo builder and a mid-sized ecommerce site will have different baselines. An honest manual audit adjusts for scale.

If you get a handful of indexing crawlers per month and zero live fetches, you have an indexing problem. Your content exists but it is not being pulled into answers. That points to a content structure problem. The fix is usually around how clearly your pages answer specific questions.

If you get live fetches but zero referral traffic, your content is being read but not chosen. That points to a trust or authority problem. The assistant read your page and decided to cite someone else. Or it cited you but the click-through rate is low because your page loads slowly or looks dated.

If you get a steady stream of both, your baseline is healthy. You can now optimize from a position of data rather than guesswork.

The distinction between SEO and AEO matters here. Traditional SEO optimizes for a search engine ranking. AEO optimizes for being cited by an AI assistant. They overlap but they are not identical. If you want the full comparison, read SEO vs AEO: do you need both in 2026?

The limits of the manual method

This manual audit has real gaps. You cannot see the prompts behind each live fetch. You cannot tell whether a citation to your page came because a user asked a specific question or because the assistant chose you from a broader set. You certainly cannot track the trends over time without repeating the whole exercise again.

Your logs tell you who visited. They do not tell you why. They do not tell you which page of yours appeared in the assistant's answer but was not clicked, because no server log records a page that was cited but never visited.

This is where the manual method ends and a dedicated tool starts. Kymo's free AI visibility checker reads a URL from public signals only. Your site does not need Kymo installed for it to work. Enter your URL, and the report arrives by email. It shows which pages AI assistants fetch and the likely prompts behind those fetches. The manual method takes an hour and gives you a snapshot. The checker does the same thing in one step.

Why this matters more than it did last year

The shift from search to answers is not hypothetical. Google still sends traffic, but a growing number of product searches start in an AI assistant. When ChatGPT answers a question, it cites a handful of sources. When Perplexity answers, it includes a source list.

If your site is not in the logs of the indexing crawlers, you are not in the candidate pool. If you are in the pool but not being cited, your content is not winning the selection. Both are fixable, but only if you know which one you are facing.

For a good plain-language explanation of the terminology, our post on what SEO for AI is called covers AEO, GEO, and AIO without the jargon. The small business guide to AEO is a deeper walkthrough if you want the full process.

Run the audit, then decide

Run an AI visibility audit by hand if you want a quick factual baseline and you are comfortable reading server logs. It is a diagnostic exercise, not a strategy session. You will find out who visits, what they read, and whether you get clicks. That is the measurement. The improvement comes after.

The honest limitation of hand-run audits is velocity. The web changes fast. Crawler registries change. New agents appear. If you run this audit once a quarter, you are flying blind in between. That is the argument for a tool that tracks continuously, not because the manual method is useless, but because it is a snapshot and visibility is a stream.

Leave the manual audit as your starting point. If it shows you are invisible, fix the indexing problem first. If it shows you are indexed but not cited, work on content structure and authority. If it shows you are cited and clicked, scale whatever you are doing.

Do the manual audit by hand if you want the baseline without installing anything. Run the free AI visibility check on your own site, or start tracking which AI engines already crawl you. The report arrives by email and reads public signals, so no installation is needed.

Start free → 14-day free trial. No card required.