Generative engine optimization (GEO) means making your business easy for AI tools like ChatGPT, Perplexity, Claude and Google's AI Overviews to find, read and cite when someone asks a question. For a small business, it comes down to three things: let the AI crawlers reach your site, keep your business facts identical everywhere, and write pages that answer real questions in the first sentence.

Most of that is ordinary SEO done well. Google's own guide to its AI features says you don't need special files, special markup or rewrites. But two parts are genuinely new. Each AI company has its own crawlers, and a site can block them without the owner knowing. And success is measured by citations in an answer, not by a position on a results page.

Below is the checklist we work through for clients, with real numbers from work we did for Elume Medspa, a med spa in Fairfax, Virginia.

Key takeaways
  • AI answer tools search the web live for current and local questions, read a few pages, and cite them. GEO is about being one of those pages.
  • Allowing the AI search bots in robots.txt isn't enough. On one client site the host was throttling GPTBot 62.5% of the time while Googlebot got through every time.
  • Google-Extended controls Gemini training and grounding. It has no effect on Google Search or AI Overviews.
  • llms.txt is optional. Google Search doesn't use it.
  • Measure by asking AI tools your customers' real questions every month and recording who gets cited.

What generative engine optimization means in plain English

When someone asks ChatGPT "how much does Morpheus8 cost in Northern Virginia," it doesn't rely only on its training. For current, local or price questions, these tools search the web, read a handful of pages, write an answer and link the sources. Google's AI Overviews and AI Mode work from Google's normal search index in a similar way.

So there are two jobs. First, your pages have to be reachable and readable by each tool's crawler. Second, your pages have to contain a clear, specific, quotable answer that the tool trusts enough to cite.

Different questions are answered from different places. In our Elume work, "med spa Fairfax" was answered mostly from the local graph: the Google Business Profile, Yelp and reviews. "How much is Morpheus8 in Northern Virginia" was answered from passages on the clinic's own pages. The first is a Google Business Profile problem. The second is fully within your control on your website.

Is GEO different from SEO?

Less than the marketing around it suggests. Google's guide to optimizing for generative AI features (published May 2026) says outright that you don't need AI text files, special schema, content "chunking" or AI-specific rewrites, and that seeking inauthentic "mentions" across the web isn't as helpful as it might seem. To appear in AI Overviews, a page just has to be indexed and eligible to show a snippet in normal Search.

What is different:

  • More crawlers. ChatGPT, Claude and Perplexity each send their own bots. Your robots.txt, firewall and host treat them separately from Googlebot.
  • Citations, not rankings. There's no position 1 in an AI answer. You're named and linked, named without a link, or absent.
  • Exact facts matter more. An AI tool that finds two different prices or two spellings of your name has a reason to cite someone else.

The GEO checklist for small businesses

Work through these in order. The first two decide whether AI tools can read your site at all.

1. Let the right AI crawlers in

Each company runs separate bots for training, for search, and for fetching a page when a user asks. They're controlled separately in robots.txt. Here's what each one does, per each company's own documentation as of October 2026:

User agentCompanyWhat it's forFollows robots.txt?
OAI-SearchBotOpenAISurfaces sites in ChatGPT search answers. Block it and you won't be shown there.Yes
GPTBotOpenAICollects content that may be used to train OpenAI's models.Yes
ChatGPT-UserOpenAIVisits a page when a ChatGPT user's action calls for it. Not used for automatic crawling.May not apply (user-initiated)
Claude-SearchBotAnthropicIndexes content to improve Claude's search results.Yes
ClaudeBotAnthropicCollects content that may be used for model training.Yes
Claude-UserAnthropicFetches a page when a Claude user asks a question.Yes
PerplexityBotPerplexitySurfaces and links sites in Perplexity search. Perplexity says it isn't used for training.Yes
Perplexity-UserPerplexityVisits a page when a user asks Perplexity a question.Generally no (user-initiated)
Google-ExtendedGoogleA robots.txt token only. Controls whether content is used to train Gemini models and for grounding in Gemini Apps and Vertex AI.Yes (it is a robots.txt rule)

Sources: OpenAI's crawler overview, Anthropic's crawler help article, Perplexity's bot guide and Google's list of common crawlers.

Google-Extended does not affect AI Overviews. Google states that Google-Extended "does not impact a site's inclusion in Google Search" and isn't a ranking signal. AI Overviews and AI Mode are built from the normal Googlebot index. If you want to limit how your content appears in them, Google's controls are nosnippet, data-nosnippet, max-snippet or noindex, and those also affect your regular search listing. Blocking Google-Extended to "stay out of AI Overviews" does nothing of the sort.

The useful decision is training versus search. You can block the training bots and still allow the search bots. A robots.txt that allows everything to do with answering questions looks like this:

User-agent: OAI-SearchBot
User-agent: ChatGPT-User
User-agent: Claude-SearchBot
User-agent: Claude-User
User-agent: PerplexityBot
Allow: /

# Training bots: your call. Allowed here.
User-agent: GPTBot
User-agent: ClaudeBot
User-agent: Google-Extended
Allow: /

On Elume's site we explicitly allow-listed all of the bots above (plus Applebot, Amazonbot, Bingbot and a few others) on August 25, 2026. For most local businesses, being read is worth more than being withheld. If you sell original writing or research, blocking training bots is a reasonable choice, and it won't remove you from ChatGPT search as long as OAI-SearchBot is allowed.

2. Check that your host isn't blocking them anyway

This is the step almost everyone skips, and it's where we found the biggest problem. robots.txt is a polite request. Your host's firewall, rate limiter or bot protection can still turn crawlers away, and nothing in WordPress or your site builder will tell you.

From our work: For Elume Medspa we pulled 72 hours of edge logs on August 25, 2026. The site's robots.txt allowed every AI bot, but the managed WordPress host's origin was returning HTTP 429 ("too many requests") selectively: GPTBot was throttled on 62.5% of requests, Amazonbot on 58.6% and ClaudeBot on 36.8%. Googlebot, Bingbot, PerplexityBot and OAI-SearchBot were throttled 0% of the time. GPTBot managed one successful content fetch in 72 hours, while chatgpt.com was the site's ninth-largest traffic source. The throttle sits at the host's origin, so fixing it needs the host's support team, not a plugin.

How to check your own site:

  1. Get your raw access logs. Most hosts offer them in the control panel or on request. If you use a CDN or firewall service, check its logs too, since it can block requests before they reach your host.
  2. Filter by each user agent in the table above and count status codes. You want 200s. Lots of 403 or 429 responses mean something is blocking.
  3. Verify the IP before you trust the name. OpenAI, Anthropic and Perplexity all publish the IP ranges their bots use. Anyone can fake a user agent.

Watch out: In the same logs, one cloud IP rotated 16 forged crawler user agents and accounted for 100% of the apparent "Google-Extended" traffic. Google-Extended never appears as a real user agent (Google's docs say crawling is done with existing Google user agents), so any request calling itself Google-Extended is fake. Also, don't test with your own fake-bot requests: an earlier reading of "harmless rate limiting" on this site came from our own synthetic probes, and we retracted it.

3. Make your business facts identical everywhere

AI tools pull your name, address, phone, hours, prices and services from your site, your Google Business Profile, directories and reviews. When those disagree, the tool either hedges or picks a source you didn't choose.

For Elume, the business name was spelled more than one way across Google, the website and the online store. On August 28, 2026 we aligned everything to one spelling, which removed 3,116 rendered occurrences of the old spellings on the website and 3,862 on the store. We deliberately left historical records and URL slugs alone, since changing a URL creates more problems than an old spelling in a slug.

Checklist:

  • One business name, character for character, on your site header, footer, schema, Google Business Profile, Yelp, Facebook and booking tool.
  • One address format and one main phone number.
  • Prices on your site that match what customers are charged. An AI answer will quote whichever number it finds.
  • Old pages with outdated prices or services either updated, redirected or noindexed.

4. Write answer-first pages

AI tools quote passages. The pages that get quoted state the answer immediately, with specifics: the price, the range, the time it takes, the town. "Our pricing varies, contact us" gives a tool nothing to cite.

  • Use the question as the heading, the way people ask it.
  • Answer in the first one or two sentences, then explain.
  • Include real numbers and say what affects them.
  • Use a comparison table when people choose between options.

Google says you don't need to break content into tiny pieces or write a special way for AI, and it's right: this is simply how a good answer page reads. For Elume we built a /questions/ page with 24 questions, including price and comparison questions, and rewrote 7 comparison posts to lead with the answer and a table.

5. Add structured data that matches the page

Structured data (schema markup) labels your business, services and prices in a machine-readable way. Google says it isn't required for its AI features and there's no special AI schema. It's still worth doing for normal search, and it removes ambiguity about who you are.

  • One Organization or LocalBusiness node with your exact name, address, phone and sameAs links to your real profiles.
  • Service or Product markup with prices that match the visible page.
  • No self-awarded star ratings, and no ratings on products that have no reviews.

Don't add FAQPage markup expecting a rich result. Google stopped showing FAQ rich results on May 7, 2026. We still use it where real Q&As appear on the page, matched word for word to the visible text, but only as a clean machine-readable copy.

Watch out: Many "AI audit" tools fetch your page as Markdown, which strips <script> tags and with them your JSON-LD schema. They then report "no schema found" on a site that has plenty. Check schema with Google's Rich Results Test or by viewing the page source.

6. Add llms.txt if you like, but don't expect much

llms.txt is a proposal from 2024: a Markdown file at yoursite.com/llms.txt with your name, a one-paragraph summary and lists of your key pages, written for AI agents to read. It's cheap to make and harmless.

Google says Google Search doesn't use llms.txt and that it will neither help nor hurt your visibility. The crawler documentation from OpenAI, Anthropic and Perplexity linked above doesn't say their bots use it either. We built a substantive one for Elume (it lists every bookable appointment link), but we put it last on this list for a reason. If your robots.txt or host is blocking bots, llms.txt changes nothing.

7. Measure citations, not just traffic

AI referral traffic is real but small for most local businesses today. Elume's site had 25 AI-referral sessions in 180 days, and the store had 24. If you judge GEO only by sessions, you'll conclude it doesn't matter, and miss whether you're being recommended at all.

A simple prompt panel works:

  1. Write 10 to 15 questions your customers actually ask: "best [service] near [town]," "how much does [service] cost in [area]," "[your business] vs [competitor]."
  2. Run each one in ChatGPT, Perplexity and Google, logged out where you can.
  3. Record for each: named and linked, named only, or absent. Note which page was cited.
  4. Repeat monthly with the same questions.

From our work: Our first prompt panel for Elume Medspa, on September 17, 2026, used 15 prompts. ChatGPT named and cited Elume on 5. Perplexity named and cited it on 4 (8 absent, 3 blocked). Google showed an AI Overview on 10 prompts and included Elume in 4. The Morpheus8 cost and location prompts were named and cited on every engine that answered, quoting the clinic's $549 and $695 prices verbatim and citing its Morpheus8 cost page. The head term "med spa Fairfax" was absent everywhere.

Two other free signals: in Google Analytics, check referrals from chatgpt.com, perplexity.ai and similar sources. In Search Console, the Generative AI performance report shows how often your links appeared in AI Overviews and AI Mode. Google says it rolled out to all sites as of August 31, 2026, though small sites may not have enough impressions to see data. It counts impressions only, not clicks.

What GEO can't do

Being honest about limits saves money:

  • It won't win broad "best X near me" answers on its own. Those lean on reviews, your Google Business Profile and how well known you are. In Elume's panel, the specific price questions were cited on every engine that answered, while the head term was absent everywhere.
  • It can't be bought with fake mentions. Google says inauthentic mentions aren't as helpful as they seem. Paying for low-quality listings and posts is the old link-scheme problem with a new label.
  • It won't fix a page nobody can crawl. If Google hasn't indexed a page, it can't appear in an AI Overview. Our SEO audit checklist covers indexing first for that reason.

If your facts are already consistent and your host isn't blocking anyone, you may not need outside help. Run the checks above, keep the monthly prompt panel going, and spend your time on pages that answer customer questions with real numbers.

Where to start this week

  1. Open yoursite.com/robots.txt and make sure OAI-SearchBot, Claude-SearchBot and PerplexityBot aren't disallowed.
  2. Ask your host for 72 hours of access logs, or a plain answer to "do you rate-limit or block GPTBot, ClaudeBot or other AI crawlers?"
  3. Search your own site for every spelling of your business name and fix the strays.
  4. Pick the three questions customers ask most and make sure one page answers each in its first sentence, with a number.
  5. Run those three questions in ChatGPT, Perplexity and Google today, and write down what you see. That's your baseline.

If you'd rather have us do it, our SEO and AI search service starts with exactly these checks, including the log review most audits leave out.

Frequently asked questions

How long does it take to show up in ChatGPT after fixing my site?

There's no published timeline. OpenAI says its systems take about 24 hours to pick up a robots.txt change, but your pages still have to be crawled and judged worth citing. We re-run the same set of customer questions about once a month and compare.

Should I block AI bots from training on my content?

That's a business choice, not an SEO one. Training bots (GPTBot, ClaudeBot, Google-Extended) are separate from the search bots that fetch pages for answers, so you can block training and still allow search. Most small local businesses gain more from being read than they lose.

Do Google AI Overviews send traffic to my website?

Sometimes. Google counts clicks from AI Overviews and AI Mode in your normal Search Console traffic, and its Generative AI performance report shows how often your links appeared there. It reports impressions only, not clicks.

Does a business need a Wikipedia or Wikidata page to show up in AI answers?

No. Those sites have notability rules most small businesses don't meet, and an entry that gets deleted helps nobody. Consistent facts on your own site, your Google Business Profile and the directories you're listed on matter more.

Will blocking AI bots hurt my Google rankings?

No. GPTBot, ClaudeBot and PerplexityBot have nothing to do with Google, and Google says Google-Extended isn't a ranking signal. Only blocking Googlebot itself would affect Google Search, including AI Overviews.