AI Guides › Step-by-step guides
By Nigel Guy · 7 min read
Most people hear that buyers now ask ChatGPT, Claude and Perplexity instead of Googling, and respond by buying a "GEO" tool, rewriting every page for robots, or adding an llms.txt file. It feels like action. Meanwhile nobody has checked the two boring things that decide whether you appear at all: whether the assistants' crawlers are allowed in, and what the assistants currently say when asked about you.
The rule: you cannot instruct a model to quote you. You can only make your pages reachable, clear and consistent, then measure what comes back.
This guide is a one-month routine built on that rule. It makes no promise of a ranking, because nobody outside the vendors controls one.
You need:
robots.txt (usually via your CMS, host or developer).Cost: nothing beyond your time. Any paid "AI visibility tracker" is optional; check the £ price at checkout and see "What to skip".
Each vendor runs separate bots for separate jobs. Blocking the wrong one is the commonest self-inflicted wound, usually from a blanket "block AI" rule copied from a forum.
| Vendor | Bot | What it does (per vendor docs) |
|---|---|---|
| OpenAI | OAI-SearchBot | Surfaces sites in ChatGPT search. Disallow it and you will not appear in ChatGPT search answers. |
| OpenAI | GPTBot | Collects content that may be used to train models. Separate from search. |
| OpenAI | ChatGPT-User | Visits pages for certain user actions; robots.txt may not apply. |
| Anthropic | Claude-SearchBot | Analyses content to improve search result quality. |
| Anthropic | Claude-User | Fetches pages when a user asks Claude for web access. Blocking can reduce your visibility to those users. |
| Anthropic | ClaudeBot | Collects web content for model training. |
| Perplexity | PerplexityBot | Surfaces and links sites in Perplexity search results; Perplexity says it is not used for training. |
| Perplexity | Perplexity-User | Fetches pages for an individual user's question; generally ignores robots.txt. |
Open yourdomain.co.uk/robots.txt in a browser. Look for Disallow: / under any of the search or user bots above, or under User-agent: *. If you want to appear in answers, allow the search bots (OAI-SearchBot, Claude-SearchBot, PerplexityBot). Whether to allow the training bots (GPTBot, ClaudeBot) is a separate business decision about your content; allowing search while blocking training is a legitimate combination.
OpenAI says changes can take about 24 hours to take effect. Also check that your host or firewall is not blocking the bots at network level, which robots.txt cannot reveal.
Write the questions your customers actually type, not your brand name repeated. Mix:
Ten is enough. More becomes admin.
Open a fresh chat each time so earlier answers do not colour the next. Ask each question as a real customer would, with web search on. Log every row in your spreadsheet. Answers vary between runs and between people, so treat one result as a sample, not a verdict.
Once the raw answers are logged, paste them into one chat with this prompt to turn them into a gap list.
You are a careful analyst helping a small UK business understand how AI assistants describe it.
Context: my business is [BUSINESS_NAME], website [DOMAIN]. I sell [WHAT_YOU_SELL] to [CUSTOMER_TYPE]. My main competitors are [COMPETITORS].
Below are logged answers from AI assistants to buyer questions, with the sources each cited.
[PASTE_SPREADSHEET_ROWS]
Task:
1. For each question, state whether I was mentioned, and whether the description was accurate, inaccurate or missing.
2. List every factual error about me, quoting the exact wording and the source it seems to come from.
3. List the sources cited most often across answers, and mark which ones I control (my site, my profiles) and which I do not (directories, review sites, press, forums).
4. List questions where competitors appeared and I did not, and name the information a page would need to contain to answer that question.
5. Finish with a ranked list of at most five fixes, each tagged "my site", "my profiles" or "third party", with the reason it ranks where it does.
Rules: use only the material I pasted. Do not invent sources, statistics or competitor details. If a log row is incomplete, ask me for it before concluding anything about that question. Before answering, check that every claim you make points to a specific row.
Fill in: your business details, competitors, and the pasted log.
Google states there are no extra requirements to appear in AI Overviews or AI Mode, no special markup, and no need for machine-readable "AI text files": normal indexing and helpful content apply. Treat that as the baseline for every assistant.
| Week | Do |
|---|---|
| 1 | Steps 1 to 3. Log the baseline. |
| 2 | Fix crawl access and publish the "what we do" page. |
| 3 | Publish question pages for your top three gaps. Correct profiles. |
| 4 | Re-run the same ten questions. Compare against the baseline. |
Re-runs after a month may show nothing. Crawling and index refresh take time, and there is no fixed schedule published by the vendors beyond OpenAI's robots.txt note. Repeat monthly.
If the answers have not moved, that is information too, not proof the work failed.
llms.txt files and "AI-only" pages. Google says they are unnecessary for its AI features. I could not confirm that the other vendors rely on them, so do not spend a week on one.