Skip to content

How to Get Your Website Mentioned by AI: A Practical Guide to GEO

8 min read5 views★ featured[Developer Workflow]
Cover image for "How to Get Your Website Mentioned by AI: A Practical Guide to GEO"

Someone asks an AI assistant which tool to use, which developer to hire, or how to solve a problem. The answer names three options and links to two sources. Your website could be useful, but it never enters the conversation.

That is the discovery problem behind generative engine optimization, usually shortened to GEO. A related term, answer engine optimization, or AEO, emphasizes appearing in answers. The terminology overlaps; the practical goal is to make your work discoverable, understandable, and credible enough to be referenced.

It is tempting to call this the replacement for SEO. I would treat it as an expansion of the job. Google explicitly says its existing SEO practices still apply to AI Overviews and AI Mode, with no separate optimization requirements. A supporting page must be indexed and eligible for a search snippet. Google's guidance for AI features

The useful shift is in what you optimize for: an answer that accurately describes your work and gives someone a reason to investigate it.

Decide what kind of visibility you want

There are three different outcomes worth separating:

  • A mention: the answer names you, your company, or your product.
  • A citation: the answer links to your page as a source.
  • A recommendation: the answer suggests you as a suitable choice for a particular need.

A tutorial can earn a citation without generating a recommendation for its author. A brand can be recommended using someone else's review, without a link to its own website. Neither outcome guarantees a visitor or a customer.

For a developer portfolio, I would pursue both useful technical articles that deserve citations and clear project pages that explain when the developer is a good fit. Those pages serve different questions.

Make the public content accessible

Start with a representative article and a project or service page. Open each without signing in. Inspect the response HTML as well as the rendered browser page. Can you find the main explanation, or only an empty application shell?

My practical preference is to deliver the important text in the initial HTML. This reduces dependence on a retrieval tool executing JavaScript successfully. It is a robustness choice, not a promise that every assistant handles pages identically.

Check redirects, broken links, accidental indexing restrictions, and firewall challenges. Use ordinary internal links so important pages are reachable from the rest of the site. Keep a current sitemap and a consistent preferred URL for each page.

For ChatGPT search specifically, OpenAI documents OAI-SearchBot as its search crawler and GPTBot as a crawler for content that may be used in model training. Their settings are independent: permitting search does not require permitting training. OpenAI also recommends allowing its published search crawler IP ranges. Official OpenAI crawler documentation

Review your existing rules before changing them. Keep private areas protected. A successful request using a copied bot user-agent does not establish what a verified crawler receives from your firewall, and a crawler visit does not establish that your page was indexed or cited.

Build pages around real decisions

Choose questions that someone with a genuine need would ask. For a portfolio, examples might include:

  • Who can build a small internal dashboard with authentication and reporting?
  • What should a founder prepare before hiring a freelance web developer?
  • How can a small team maintain a custom application after launch?

Use those questions to identify missing information, rather than repeating them mechanically throughout the page.

Consider this fictional service description:

I build innovative digital experiences using modern technologies.

It offers little evidence for deciding whether the developer fits a task. A more useful description would be:

I build internal dashboards for small operations teams, including authentication, reporting, and data imports. Project pages explain the requirements, implementation choices, and handover process.

The second description identifies an audience, a deliverable, and a path to verification. Use that level of specificity only where it accurately describes your work. If your scope is broader, organize it into clearly explained services instead of inventing a specialty.

Give each important claim something to stand on

My recommendation is to prioritize evidence you can publish from actual work. A generic article about performance competes with countless similar explanations. A documented investigation of a specific slow page gives readers something more concrete to evaluate.

A strong project write-up can explain:

  1. The original problem and constraints.
  2. Your role and the decisions you made.
  3. The implementation, with screenshots or a public demo where appropriate.
  4. The outcome and how you evaluated it.
  5. The limitations and what you would change next time.

If you report a performance improvement, include the test conditions and distinguish laboratory measurements from real-user data. If you cannot disclose a client, explain what is anonymized. If a project is a personal experiment, say so.

This is an editorial strategy, not a proven universal ranking formula. It gives both readers and systems that retrieve your page more substance to work with. Invented numbers and unsupported claims would undermine the entire purpose.

Write passages that survive being read alone

An explanation is easier to reuse accurately when its main point and qualifications appear together. Lead a section with the answer, then explain the reasoning and examples.

For example, a tutorial might say:

A database index can reduce the rows a query needs to inspect, but it adds storage and write overhead. Check the query plan against realistic data before deciding that a new index solves the problem.

That passage states the mechanism, tradeoff, and next step without requiring the reader to reconstruct several earlier paragraphs.

Use descriptive headings. Name the tool or project when a vague “it” would be ambiguous. Keep dates, units, assumptions, and exceptions close to the claims they qualify. Add a short FAQ only when it answers questions the main article genuinely leaves open.

There is no reason to force every paragraph into a fixed length or every page into an FAQ template. The aim is an explanation someone can understand and verify.

Make your identity easy to verify

For a personal site, connect the About page, project pages, articles, and relevant public profiles. Use a consistent name and describe your role accurately. Explain who wrote an article and how their experience relates to the topic.

External evidence is also worth developing through normal professional work: public repositories, useful contributions, published talks, and client case studies you have permission to share. I would pursue these because they let other people inspect your work independently; I would not assign them an invented AI ranking weight.

If another site describes your project inaccurately, request a correction. Avoid buying fabricated reviews or manufacturing endorsements. A recommendation based on false information creates expectations you cannot meet.

Treat markup as supporting infrastructure

Structured data should accurately describe the content already visible to visitors. It cannot substitute for an informative page.

Google explicitly says its AI search features require neither special AI text files nor a special schema.org type. That includes no requirement to add an llms.txt file for those features. This statement concerns Google's documented search features, not every possible AI tool. Google's technical guidance

I would spend the first hour fixing inaccessible content or an unclear project description before adding another machine-readable summary. If you do provide supplementary files, maintain them alongside the pages they describe.

Measure discovery, accuracy, and useful visits separately

Create a small, repeatable set of questions based on your audience's needs. Include questions that do not mention your name; otherwise, you are mostly checking whether an assistant can find an entity you already supplied.

For each observation, record the question, date, assistant, search mode, whether you appeared, the linked source, and whether the description was accurate. Repeat observations over time. Treat this as a sample of behavior, not a global visibility score or a controlled proof that one edit caused an improvement.

Bing's AI Performance report provides citation information for supported Microsoft and partner experiences. Its documentation distinguishes citation counts from traffic and warns against attributing changes to a particular cause. It does not measure every assistant. Bing AI Performance documentation

Then examine referrals and meaningful actions on your own site: project inquiries, demo visits, or newsletter subscriptions. A citation with no click may still create awareness, but you should not report it as a visit. Where attribution is unclear, a simple “How did you hear about me?” field can add useful context.

Start with one page worth recommending

For a first pass, choose one important project or service page and one supporting article. Verify access, make the audience and scope explicit, add evidence, and connect the pages with useful links. Record your baseline questions before making changes, then revisit them after allowing time for discovery.

You cannot configure an assistant to recommend you whenever someone asks. You can publish a site that clearly explains what you do, demonstrates it honestly, and answers the questions someone would ask before choosing you. That is a practical foundation for earning both citations and the right recommendations.

# Comments (0)

No comments yet. Be the first.

0/1000