GEO Audit Checklist: 7 Checks to See If AI Can Find and Recommend Your Business

A customer asks ChatGPT for a good emergency plumber in Warrington and gets three names. Yours isn’t one of them. And you’ve no way of knowing whether that’s because the AI can’t read your site, doesn’t trust your details, or simply picked someone else that time.

That’s what a GEO audit checklist is for. A GEO (generative engine optimisation) audit checks whether AI assistants can reach your website, understand what you do and where, and find enough agreement elsewhere online to recommend you. If the term is new, start with what GEO actually is.

It’s less mysterious than it sounds. Google’s own guidance says optimising for its AI features is “still SEO”. So this AI search visibility audit only covers what a normal local SEO audit misses: crawlers other than Google’s, the outside sites assistants draw on, and testing more than once.

There are seven checks, in order. Diagnose now, fix later: write every failure down as you go.

Step 1: Ask the AI tools what your customers ask, and ask more than once

SparkToro and Gumshoe ran 2,961 prompts through ChatGPT, Claude and Google’s AI Overviews. For ChatGPT and Google’s AI, the chance of any two responses giving the same list of brands was less than 1 in 100. One screenshot tells you almost nothing. Your goal is a written baseline you can re-run next month.

1. Write 8 to 10 questions the way a customer would ask them. Mix the types:
   • Service and town: “emergency roofer Widnes”
   • Near me: “dentist near me open Saturday”
   • Problem-led: “mice in my loft, who should I call in St Helens?”
   • Comparison: “best family solicitor in Warrington for divorce”
   • Your name: “what do people say about [your business name]?”
2. Run each one in ChatGPT, Gemini, Perplexity and Google (AI Overview or AI Mode, where one appears). Use a fresh chat, logged out if you can.
3. Ask local questions both ways. OpenAI says ChatGPT may rewrite “near me” to include the town it thinks you’re in, so ask both.
4. Run the whole set at least three times, on different days.
5. Log every run: named (yes/no), linked (yes/no), what it said about you and whether it was accurate, and which competitors appeared.

You should end up with an “appears in X of Y runs” figure per question, plus the competitors who keep turning up: your real AI competitor set. Unsure it matters for your trade? Read how AI search affects local businesses.

Paid trackers such as Profound and Peec AI automate this, but they’re built for enterprise budgets. If you only have 20 minutes, run your three money questions in two tools, twice.

Step 2: Check which AI crawlers your site actually lets in

Blocking GPTBot doesn’t keep you out of ChatGPT search answers. Blocking OAI-SearchBot does. Most checklists don’t tell them apart. To audit your website for AI search, open yoursite.co.uk/robots.txt and search it for each name below.

Google-Extended is the odd one. Google says it “does not impact a site’s inclusion in Google Search”, but it also controls grounding in Gemini Apps, not just training. Blocking it is a trade-off if you care about the Gemini app.

Then check your CDN, because robots.txt can look clean while a firewall blocks the bots. Since 1 July 2025, Cloudflare has asked every new domain whether it wants to allow AI crawlers. For new domains from 15 September 2026, it blocks training and agent crawlers by default on pages that display ads, while search crawlers stay allowed. On Cloudflare? Check the AI crawler setting in the dashboard.

By now every search crawler should be allowed, and any training block should be a deliberate choice. OpenAI says changes take about 24 hours to register.

Step 3: Check your key pages still make sense with JavaScript switched off

Your site can look perfect in your browser and be half empty to ChatGPT.

In December 2024, Vercel and MERJ found “none of the major AI crawlers currently render JavaScript”, including OAI-SearchBot, GPTBot, ClaudeBot and PerplexityBot. Google says it can process JavaScript as long as it isn’t blocked. So doing well on Google isn’t proof the others can read you.

The goal is to confirm your key facts sit in the page’s raw HTML. Pick one method:

1. Right-click the page, choose View page source and press Ctrl+F. Search for your phone number, one service and one town.
2. Or turn JavaScript off in Chrome (Settings > Privacy and security > Site settings > JavaScript) and reload.

Check your homepage, top service page and top location page for services, towns covered, phone number, opening hours, prices and reviews. The usual culprits are review widgets, script-built accordions and tabs, and text inside images.

By now you should know whether your key facts appear in the source of all three pages.

Best for: page-builder and app-style sites. Skip if: view-source already shows everything on a simple WordPress theme. Either way, the fix is a developer job. Note it and move on.

Step 4: Check your service and location pages answer the question in plain text

Read the first two sentences of your main service page out loud. If they don’t say what you do and where, an AI assistant has nothing clean to quote.

The goal is a quotable answer near the top of every page that earns you money. Check your main service page and one location page:

• The opening states service, area and customer. For example: “We repair and replace flat roofs across Widnes, Runcorn and Warrington, for homes and small commercial buildings.” Not “Welcome to our family-run business”.
• Practical facts are in the text. Areas covered, availability or response time, a “from” price or how pricing works, accreditations. Only what’s true.
• Headings match how people ask. A question or a plain statement beats a clever slogan.
• Location pages say something specific to that town. Swapping the place name gives an assistant no reason to prefer the page.

This isn’t rewriting for robots. Google says “there’s no requirement to break your content into tiny pieces for AI to better understand it.” Stating facts clearly and early helps human readers just as much.

One schema sub-check: if you have LocalBusiness or other structured data, Google says it should match the visible text. Wrong opening hours in your schema is a fail. No schema isn’t.

By now you should have a sentence on each page you’d be happy to see quoted word for word.

Step 5: Check your business facts match everywhere AI assistants look

Go back to your Step 1 log. Did any assistant give an old phone number, wrong hours, or a town you stopped covering? That came from somewhere.

AI answers about local businesses draw on listings data as well as your website. OpenAI says ChatGPT may use trusted third-party providers for local information such as nearby business listings. Google says Business Profiles can help your services appear in AI responses, and advises keeping that information up to date. Its guidelines also ask you to represent the business as it’s consistently known in the real world.

The goal is one set of facts, everywhere.

1. Write a master record: name, address or service area, phone, website URL, opening hours, main services and towns covered.
2. Compare it against your website footer and contact page, Google Business Profile, Bing Places, Apple Business Connect and the main UK directories you’re on. For the full directory list and how to check name, address and phone properly, use step 4 of our local SEO audit.
3. Compare services and areas, not just contact details. If your Business Profile leaves out a service you want to be recommended for, that’s a gap.

You should finish with a list of every mismatch. If you’ve moved or changed number in the last three years, start here.

Step 6: Check what your reviews and other websites say about you

Look at the sources Perplexity and Google linked beneath their answers in your Step 1 log. That’s a free list of the websites deciding who gets recommended in your trade and town, and your goal is to find out whether you’re on them.

1. List every domain cited: directories, review sites, local news, trade bodies. Note where competitors appear and you don’t. See how AI search citations work for why these matter.
2. Expect each assistant to lean on different sources. A Yext study (Yext sells listings) found Gemini cited brand websites most, ChatGPT leaned on third-party listing and review sites, and Perplexity on niche directories. It’s vendor research, but the lesson holds: one strong source isn’t enough.
3. Read your last 10 reviews. Do customers mention the service and the town in their own words? Reviews are part of the Business Profile Google says can help you appear in AI responses, so substance matters, not just stars.
4. Check the Step 1 answers to “what do people say about [your business name]?” Are they fair and current?

Don’t manufacture any of this. Google says “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem.” Fake reviews and paid listicle spam aren’t the fix.

You should end with three to five sources you should be on and aren’t.

Best for: businesses with a decent rating but thin, generic reviews. Skip if: Steps 2 to 5 aren’t done yet.

Step 7: Score your results, fix in order and re-test the same way each month

A list of failures isn’t a plan. Some fixes are pointless until others are done, so work in this order:

1. Blocked search crawlers or CDN blocks. Often an hour’s work, and nothing else counts until it’s done.
2. Content hidden behind JavaScript.
3. Conflicting business facts.
4. Service and location pages with no plain answer.
5. Reviews and third-party presence. The slowest, and never really finished.

Then re-run the same questions, in the same tools, the same number of times, around the same date each month. Track “appears in X of Y runs” and whether descriptions are accurate. Never compare one run with one run: as Step 1 showed, answers barely repeat.

For Google, add Search Console’s Generative AI performance report: impressions (not clicks) from AI Overviews and AI Mode, by page. Google began testing it with some UK sites in June 2026 and says it reached all websites by 31 August 2026.

Two things aren’t fails: no llms.txt file and no special “AI schema”. Google says you need neither.

If your list runs long, or the fixes need a developer, tell us what you do and where. We’ll give you a straight answer on whether we can help. If we can’t, we’ll say so.

FAQ

Do I need an llms.txt file for AI search?

No. Google says you don’t need new machine-readable files or AI text files to appear in its search, including its AI features. Google’s John Mueller said in June 2025: “no AI system currently uses llms.txt.” OpenAI, Anthropic and Perplexity all point site owners to robots.txt to control their crawlers. Adding an llms.txt file does no harm. Not having one isn’t an audit failure.

Will blocking GPTBot or Google-Extended remove me from AI answers?

Not from search answers, but there’s a catch. GPTBot is OpenAI’s training crawler; ChatGPT search uses OAI-SearchBot. Google-Extended doesn’t affect Google Search or AI Overviews, but it does control grounding in the Gemini app. Block training crawlers if you want to, and keep the search crawlers open. See Step 2 above for the full list.

How is a GEO audit different from a local SEO audit?

Most of the foundations overlap. Google itself calls optimising for its AI features “still SEO”. A generative engine optimisation audit adds what a local SEO audit skips: non-Google crawlers, whether they can see content built with JavaScript, the outside sources assistants draw on, and repeated prompt testing. If you’ve never run a local SEO audit, do that first.

How often should I re-run the checks?

Re-test your question set monthly. Re-check robots.txt and CDN settings whenever your site, host or security plugin changes, and at least twice a year. Check business facts quarterly, or straight away after moving, rebranding or changing number. OpenAI says robots.txt changes take about 24 hours to register, but being recommended takes longer and isn’t guaranteed.