Technical GEO: a website AI assistants can read and quote

Generative engine optimization is the work that lets ChatGPT, Perplexity, Google AI Overviews, Claude and Copilot reach your pages, understand who you are and quote you with the right facts. We check what these assistants actually see and say about you, fix the site, and ask the same questions again afterwards.

When GEO is worth doing

More buyers now ask an AI assistant before they open a search results page. If the assistant cannot read your site, it answers from whatever else it finds.

Competitors come up when people ask ChatGPT about your field, and your company does not
An AI assistant describes you with an old address, a former name or somebody else's services
Nobody on the team knows whether the firewall or robots.txt lets AI crawlers in
Prices, contacts or key text load through JavaScript, and a crawler sees an empty page
The site runs in several languages, but AI answers only reflect the English version

What we do

AI crawler access

We send real requests as GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and other AI crawlers and see what comes back. robots.txt, Cloudflare bot settings and WAF rules often block them without anyone deciding to.

Content without JavaScript

Most AI crawlers do not run scripts. We make the main text, prices, contacts and FAQ part of the HTML the server sends, with server rendering where the site is a JavaScript app, and keep e-mail addresses out of script-based obfuscation.

llms.txt

A short file at the root of the site with the company facts, the main services and links to the pages that answer common questions. It is written from the same facts as the site and updated when they change.

Company entity and schema.org

Organization markup with the legal name, founding year, addresses and sameAs profiles, plus Service, FAQPage, Article and BreadcrumbList, all linked by @id. The facts in the markup match the text on the page.

Text that can be quoted

Assistants quote short passages. We rework key pages so each section starts with a direct answer, figures come with a date or a source, comparisons sit in tables and FAQ answers the questions buyers really ask.

Answer checks before and after

We agree a set of questions your buyers ask, put them to ChatGPT, Perplexity, Google AI Overviews, Claude and Copilot, and record who gets mentioned, what is said about you and which pages are cited. After the fixes we ask again.

First question: can AI assistants reach your pages at all?

Before any content work we check the plumbing, because it is where most problems are. A bot rule added against scrapers years ago may also stop the crawlers that ChatGPT search or Perplexity use. A single-page app may send crawlers a shell with a loading spinner. A CDN may replace your e-mail address with a script, so an assistant reading the page never sees it. Each of these is a small fix once someone knows about it, and each one silently removes you from the answers until then.

A list of AI crawlers with what each one gets from your site today
A decision per crawler: allowed for search, blocked for training, or both
The key pages checked as a crawler sees them, without JavaScript
robots.txt, CDN and firewall rules changed to match that decision

One set of facts, the same in every language

Assistants build their picture of a company from many pages at once. When one page says fourteen years, another says fifteen and an old profile still carries a former brand name, the answer comes out vague or wrong. We did this work on our own site first: 772 URLs in ten languages. It now has an llms.txt with the company facts and every service, FAQPage markup on all service pages in all languages, Organization markup with the legal name and founding year and without the old brand’s profiles, and pages, services and FAQ linked by @id. AI crawlers get a normal response, and e-mail addresses are in the HTML instead of behind a CDN script.

A fact sheet: name, legal entity, founding year, locations, services, contacts
The same facts in the text, the markup and llms.txt, in every language
hreflang so each language version is found and credited
Old names and outdated profiles removed from sameAs and page text

Measured on the questions your buyers ask

Nobody can see inside an AI model, so we measure from the outside. Together we write a set of questions: the category with your city or country, comparisons with competitors, typical problems your service solves, and your company name. We put them to several assistants, save the answers and the cited sources, and repeat the same set after the fixes and later on a schedule. Answers vary from one run to the next, so the report looks at patterns across the set rather than a single reply.

How the work goes

Questions and baseline

We agree the question set and record today’s answers from the main AI assistants. You see where you are mentioned, where competitors are and what is said about you.

Access check

We test AI crawler access, rendering without JavaScript, robots.txt, CDN and firewall rules. You get a list of what blocks or hides your content.

Facts and markup review

We compare company facts across pages, languages, markup and profiles, and review schema.org and llms.txt. You get a fact sheet and the list of mismatches.

Fixes

Our developers change the site, markup and files, and rework the key texts, or we brief your team and review their changes.

Re-check

Once crawlers have revisited the pages, we run the same question set again and compare the answers with the baseline.

Monitoring

If you want, we repeat the checks on a schedule and after major releases, when crawler access and markup most often break.

Crawlers, standards and tools

AI crawlers and tokens

Markup and files

Checks

Platforms

Ways to work with us

GEO audit

A fixed-scope audit: baseline answers, crawler access, facts and markup, with a report ordered by impact. Your team makes the fixes.

Audit and fixes

The audit, then the changes by our developers, then the same question set again to compare with the baseline.

Ongoing checks

Repeat answer checks and access tests on a schedule, often combined with technical SEO or website support.

Reviews

We treat each client and his project with love.

Revol team continues to streamline the client’s development capabilities through their high-quality work and reliable support. They communicate effectively and exhibit a strong understanding of the client’s needs and business.

Tomas

CEO (Yukon, Canada)

The work from Revol, have fully met expectations and satisfied the client. Their fresh approach and ability to remain available for support have been valuable assets. Customers can bring them on for a communicative, client-oriented team to achieve their goals with.

Andrey

CEO, Stellium (USA)

Frequently asked questions

Can you guarantee that ChatGPT or Perplexity will recommend us?

No. Nobody outside these companies controls what an assistant says, and answers change between runs and model versions. We can make sure your site is reachable, your facts are consistent and your pages are easy to quote, and we show the answers before and after so you can judge the result yourself.

How is this different from technical SEO?

Technical SEO is about search engines: crawling, indexing, speed, redirects and rankings in the results page. GEO is about AI assistants that read your pages and write an answer: crawler access, content without JavaScript, company facts, markup and passages they can quote. The two share groundwork, and many sites need both, so we often do them together.

Should we allow AI crawlers or block them?

It is a business decision, and it can differ per crawler. Some crawlers fetch pages for live answers and cite the source, others collect data for model training. Many companies allow the first and block the second. What we see most often is neither: crawlers blocked by accident through a firewall or CDN setting nobody remembers.

How soon will AI answers change after the fixes?

Assistants that search the web while answering, such as Perplexity, ChatGPT search and Google AI Overviews, can pick up changes once their crawlers revisit the pages, usually within weeks. What a model knows without searching changes only with a new model version, which nobody can schedule. That is why the re-check comes after a pause, not the next day.

Does llms.txt make AI assistants read our site?

Not by itself. Support for it differs between assistants, and most do not say whether they use it. It costs little and gives a clean summary of your facts, so we add it, but crawler access and the content of the pages matter much more.

Our site is in several languages. Why do AI answers only use the English version?

Usually because the other versions are harder to find or say different things: missing hreflang, translated pages without markup, older facts in some languages, or language versions that load through JavaScript. We check every language against the same fact sheet and the same markup.

Do you write the content?

We rework existing key pages so they answer questions directly and carry checkable facts, and we tell you which facts are missing, such as results in case studies. We do not produce articles in volume or build links.

What access do you need?

For the audit, only the public site, plus Search Console and Bing Webmaster Tools if you have them. For fixes, access to the CMS or the repository and to the CDN or firewall settings. Everything is agreed in advance.