The GEO and AEO checklist: what to do, and how strong the evidence is

Most GEO and AEO checklists present every tip as equally proven. They aren’t. Some steps are documented by the platforms, some are backed by research, and some are informed guesses. This checklist tells you which is which, so you can spend your time on the work most likely to count.

How to read the evidence labels

Each section below carries one of three labels.

Label	What it means
Platform confirmed	Google, OpenAI, Microsoft or Perplexity documents it.
Research backed	A published study supports it. Often a correlation, not proof of cause.
Our judgment	Sensible and low risk, but not proven. Test it and see.

Your progress on each list saves in your browser, so you can work through this over a few sittings.

1. Let assistants read your site

Evidence: platform confirmed. Google lists crawl access in robots.txt and in your CDN or hosting setup as a best practice, and says important content should be in text form. OpenAI says a site must not block OAI-SearchBot to appear in ChatGPT’s summaries and snippets.

Access

0 of 6 done

• robots.txt allows Googlebot, Bingbot, OAI-SearchBot and PerplexityBot
• Your CDN, firewall or security plugin isn’t blocking those crawlers
• You’ve decided on GPTBot separately (it relates to AI training, not ChatGPT search)
• Your key pages are indexed in both Google and Bing
• Your services and service area appear as text, not only in images or PDFs
• Your sitemap is submitted to Google Search Console and Bing Webmaster Tools

The firewall check is the one people miss, because robots.txt can look fine while a security setting blocks the crawler. Here’s a quick test:

curl -s -o /dev/null -w "%{http_code}\n" -A "OAI-SearchBot/1.0; +https://openai.com/searchbot" https://yoursite.com/

A 200 is a good sign, but not proof: this only copies the crawler’s name, and some firewalls also check where a request comes from. A 403 means something is probably blocking it. If you can, check your firewall logs too.

2. Say plainly what you do, where and for whom

Evidence: platform confirmed, plus what we saw. Bing recommends clear headings, tables and FAQ sections. In our own test of agency questions, Google’s AI Mode put first an agency whose site names ChatGPT, Gemini and Claude outright. That’s one example, not proof, but assistants often repeat how a business describes itself.

Clarity

0 of 6 done

• Every service has its own page
• Each page opens with a sentence that could be quoted alone: who, what, where
• The towns or industries you serve are named in plain words
• Service pages answer real buyer questions in a short FAQ
• Headings and tables are used where they make a page easier to scan
• If you offer it, you name the exact thing: “AI search optimization for Google AI Mode and ChatGPT”, not “future-ready marketing”

3. Give assistants evidence worth quoting

Evidence: research backed and platform confirmed. The Princeton-led GEO study found that adding citations, quotations and statistics raised visibility by more than 40% in its tests. Bing also advises backing up claims with data, examples and sources, and keeping content current.

Evidence

0 of 6 done

• Claims have a specific number, example or named source behind them
• Sources are linked, so a reader or assistant can check them
• Pages show a real published or updated date
• Old pages get reviewed and corrected, not just re-dated
• You publish something others can’t copy: your own data, tests or prices
• Case studies name the client and show the results, with permission

Original information is the strongest item on that list, and the hardest to fake. If ten sites say the same thing, an assistant can quote any of them. If you’re the only source for a number, it has to come to you.

4. Add structured data that matches the page

Evidence: platform confirmed, with a limit. Google says no special schema is required for its AI features, and that structured data must match the visible text. It’s still useful for stating your facts clearly, but it won’t make up for weak pages.

Structured data

0 of 5 done

• Organization or LocalBusiness markup with your name, phone and service area
• Service markup on each service page
• FAQ markup only where the questions and answers are visible on the page
• sameAs links to your Google Business Profile and social profiles
• Everything in the markup matches what the page says

5. Fix the listings assistants read

Evidence: research backed and platform confirmed. BrightLocal’s 2026 study of about 200,000 local AI searches found business websites made up 42% of citations, Google Business Profile 28.6% and Yelp 9.5%. It also found ChatGPT relies more on Yelp and Bing and hardly uses Google Business Profile. So a strong Google profile alone won’t cover ChatGPT. Bing tells local businesses to keep Bing Places current.

Listings (for local businesses)

0 of 7 done

• Google Business Profile is complete, with services and real photos
• Bing Places is claimed and matches your Google profile
• Yelp is claimed and complete
• Apple Business Connect is claimed
• LinkedIn and Facebook company pages are complete
• You’re listed on the directories your industry actually uses
• Name, phone and description match everywhere

6. Earn mentions on other sites

Evidence: research backed. Ahrefs studied 75,000 brands and found branded web mentions correlated with AI Overview visibility far more strongly than backlinks (0.664 vs 0.218). That’s a correlation, and well-known brands get both. But it’s the best reason to think beyond your own website.

Reputation

0 of 5 done

• A steady flow of genuine reviews, on Google and elsewhere
• Listed by your chamber of commerce or trade association
• Included on independent “best of” lists that already rank
• Local press or trade coverage, ideally around your own data
• Real participation where your buyers ask questions, with your affiliation stated

A note on lists you publish yourself: Peec’s analysis of 232,000 citations found self-promotional lists make up about 11% of AI citations, but only 3.6% on ChatGPT. The authors still advise against them because of reputational and penalty risk. Being included on someone else’s list is the safer goal.

7. Measure it properly

Evidence: platform confirmed and research backed. BrightLocal found only 20–33% overlap between repeat runs of the same question, so one check tells you very little.

Measurement

0 of 7 done

• A fixed list of questions your buyers would really ask
• Each question asked several times on each assistant, every month
• Results logged as a share of answers, not just yes or no
• Bing’s AI Performance report checked for citations and cited pages
• Search Console reviewed (Google includes AI Mode and AI Overviews traffic in its Web totals)
• Analytics checked for AI referrals, such as utm_source=chatgpt.com
• New leads asked how they heard about you

We call the main number Recommendation Share:

Recommendation Share = answers naming your business ÷ total answers

Always show the count beside the percent. “4 of 15 answers” is honest. “27%” on its own can hide a tiny sample.

What we’d skip for now

• llms.txt and other AI text files. Google says you don’t need new machine-readable files or AI text files for its AI features. Other assistants may read them, but that alone isn’t a reason to prioritize one.
• Rewriting everything “for AI”. Google says its usual advice for helpful content and technical basics applies.
• Hidden text or instructions aimed at crawlers. It’s manipulative, and it isn’t worth the risk to your reputation.
• Fake or incentivized reviews. Google’s policy prohibits fake engagement.
• Anyone who guarantees a recommendation. Nobody controls what assistants say.

Our own starting point

A checklist is easy to write, so here’s where we stand. On November 3, 2026, we ran a small pilot on Google AI Mode: 10 questions about agencies in Iowa City, Cedar Rapids and Iowa, for 19 usable answers. Leadsure was named in 0 of 19. Our website was only days old, and this was one assistant, so it’s a starting line and not a finding.

The pilot still showed us a few things:

• The agencies named mostly had about 10–70 Google reviews, not hundreds.
• Bing’s AI answer named a different set of agencies than Google for the same kind of question.
• One Google answer cited a competitor’s LinkedIn company page as a source.

We’re working through this list on our own site and will update this post with what moves and what doesn’t.

Frequently asked questions

What’s the difference between GEO and AEO?

Very little in practice. AEO (answer engine optimization) came first and focused on direct answers. GEO (generative engine optimization) came from the research on AI-generated answers. Both describe the work on this page.

Do I need separate content for each AI assistant?

No. The same site serves all of them. What differs is the outside sources they lean on, which is why listings beyond Google matter.

How long does it take to show up in AI answers?

There’s no reliable number. Access and clarity fixes can be picked up as soon as pages are re-read, while reviews and mentions build over months. That’s why it’s worth logging results each month from day one.

Can I do this myself?

Yes. Most of this is careful work, not special tools. If you’d rather have it done, or want a second opinion, our free audit includes what AI assistants say about your business today.