Insights · GEO

Make your business legible to AI search: the exact checklist we run.

Romaan Sheikh, Director · August 2026 · 8 min read

In one 24-hour window in mid-August 2026, 37 requests from AI crawlers hit this website: Claude-SearchBot alone made 24, and the most-requested file was our sitemap. When someone asks ChatGPT or Claude "who does marketing for fostering agencies?", the answer is assembled from what those crawlers could read. GEO, generative engine optimisation, is the work of making sure they read you correctly. Here is the actual checklist we run, including the trap we caught on our own domain.

First, the trap we fell into ourselves

We audited our own site and found our robots.txt was blocking GPTBot, ClaudeBot, Google-Extended and CCBot, the crawlers for ChatGPT, Claude and Gemini. We hadn't written a single line of it. Our CDN had a "managed robots.txt" feature switched on, silently rewriting the file to block AI crawlers as a default.

A GEO agency blocking AI crawlers is an embarrassing confession, and we're making it deliberately: if it happened to us, it is happening to you. Check yours in thirty seconds: open yourdomain.com/robots.txt and look for GPTBot, ClaudeBot or Google-Extended followed by Disallow: /. If they're there and you didn't choose that, your CDN or security plugin chose it for you.

The checklist

1 · Crawler access (2 min). As above. The decision has a genuine trade-off: allowing crawlers means your content feeds AI answers, sometimes without attribution, but for most businesses, absence from AI answers costs more than inclusion does.

2 · One page that answers the question (the real work). AI engines cite pages that answer questions directly and concretely. "Award-winning solutions for ambitious brands" is illegible to a language model and human alike. "Fixed-price websites: £299 for one page, £999 with booking built in, care included" is quotable. Audit your key pages: could a machine extract what you do, for whom, at what price, from the visible text alone?

3 · Structured data that matches reality (1–2 hrs). Schema.org markup is how you state facts unambiguously: Organization with legal name and address, Service with real prices, FAQPage for genuine questions. Ours declares the company number and ICO registration, facts an AI can verify against Companies House. Validate what you have before adding more; invalid schema is worse than none.

4 · llms.txt (30 min, speculative). A plain-text summary of your business at /llms.txt for AI consumption. Honest caveat: adoption by the engines is unproven. We publish one because it costs half an hour and states our facts in exactly the form a model wants them, but it goes at the end of the list, not the start.

5 · Entity corroboration (1 hr). AI engines cross-reference. Your name, address and details should read identically on your site, Companies House, Google Business Profile and social profiles, tied together with sameAs markup. Inconsistency reads as uncertainty, and uncertain facts don't get cited.

6 · Speed and stability (ongoing). Crawlers have budgets. A slow site gets read less often and less deeply. This site holds Core Web Vitals green with zero layout shift, not vanity, crawl efficiency.

7 · Watch the crawlers arrive (monthly). Your CDN or server logs show which AI crawlers visit and what they read. That 37-requests figure at the top of this page is from our own dashboard, it's also your evidence of whether any of this is working.

What GEO cannot do

Honesty clause: nobody can guarantee your business appears in ChatGPT's answers. Anyone promising that is selling something broken. Citation follows retrieval indexes and training cycles that lag months behind your changes, and the engines decide, not you. What you control is legibility: whether, when the engines look, they find clear facts, real numbers and a consistent entity. That part is entirely mechanical, hence a checklist.