SEO and GEO: visibility in search and AI.

01 /How AI answers change search visibility

Classic search usually offers a ranked set of pages. A generative result may instead assemble an answer from several retrieved sources and place links beside particular claims. That changes the shape of visibility: a page can contribute to an answer even when the visitor never sees a familiar list of ten blue links, and one broad question may trigger several narrower retrieval queries behind the scenes.

The underlying requirements did not disappear. Google states that its generative search features use its search index and core ranking systems. A page still needs to be crawlable, indexable, relevant and useful. For Google, there is no separate GEO markup or special technical gateway. Other answer engines have their own crawlers and controls, but none can cite a page they cannot retrieve or understand.

GEO adds a question to ordinary SEO work: can this page support part of an answer with clear, verifiable information? It cannot compensate for repetitive articles that offer little useful detail. Focus on the specific explanation, evidence or experience the reader needs.

02 /Crawling, indexing and training are separate choices

Begin with the ordinary technical path. The canonical URL should return a real 200 response, expose the main content in usable HTML, allow the intended crawler and avoid a noindex directive. Internal links and a sitemap help discovery; canonical and hreflang signals help systems understand equivalent pages. Rendering the entire answer only after a fragile client-side request adds risk without making the content more modern.

Do not reduce every bot decision to 'allow AI' or 'block AI'. OpenAI documents OAI-SearchBot for inclusion in ChatGPT search, while GPTBot controls potential training use. Google uses Googlebot and normal Search preview controls for its AI search features; Google-Extended applies to some other AI uses, not ranking in Search. A publisher can therefore make different decisions about search discovery, snippets and model training.

Write those decisions down and test the delivered response. A robots.txt file can be syntactically correct yet contradicted by a CDN rule, a staging header or a firewall challenge. Fetch important URLs with the relevant user agent, inspect the rendered HTML and confirm that canonical, indexability and status codes agree.

Also check the generative-search inclusion control in Search Console and any inherited parent-property setting. This is separate from robots.txt and training choices.

• Google · Search generative AI control ↗

For an access or indexing problem, follow the step-by-step indexing diagnosis.

03 /Write useful explanations supported by evidence

A useful guide helps the reader make a decision. It can document a test, explain a commonly confused distinction, show the limits of an approach or define a small pilot. Answer the main question early, then add the detail needed to assess it.

Bring evidence that belongs to the author: a repeatable method, a photograph of the actual kind of work, an original diagram, a clearly labelled fictional example, or a technical distinction learned through implementation. Link consequential claims to primary sources. This does not mean inserting a citation after every sentence; it means making it easy to tell observation, recommendation and external fact apart.

One focused article can cover several natural formulations of the same problem. It does not need cloned pages for every long-tail phrase. Google explicitly warns against creating large numbers of query variants for manipulation, and its systems understand related wording. Use the vocabulary readers use in titles and headings, then write the body for the decision rather than repeating the keyword.

• State the practical answer before the history of the topic.
• Include a boundary: when the recommendation is wrong or insufficient.
• Use real experience or mark constructed examples honestly.
• Cite primary documentation for claims that can change.
• Remove repetition and add specific examples or evidence where an explanation is too generic.

04 /Make the page easy to inspect, quote and navigate

Clear structure helps people first. Use one descriptive title, one main heading and sections whose headings say what they resolve. Tables are good for exact comparisons; lists are good for checks; prose is better for caveats and reasoning. Tiny artificial 'answer chunks' are not required. A section should be as long as its idea needs and no longer.

Keep key statements in text, even when a photograph or diagram adds understanding. Give useful images descriptive alt text and meaningful filenames, but do not turn the alt attribute into a keyword list. Link related guides where the next question genuinely follows, and link the relevant service after the reader has enough information to judge it.

Structured data should describe visible content: an Article with its author and dates, a BreadcrumbList matching navigation, or a Service describing the actual offer. It can clarify information and support eligible search features. There is no special GEO schema, and markup must not invent reviews, FAQs or locations.

05 /Create location pages when the location changes the answer

International visibility starts with language that feels native and an offer that works operationally in that market. A German page should use the words a German buyer uses, answer the contractual or process concerns that matter there and state how communication works. An English page can serve the United States and United Kingdom when the problem is universal; tax, regulation or procurement topics may deserve a country-specific article.

Create a location page only when the location changes the answer: a real office, local service area, on-site availability, regulation or a body of place-specific work. Swapping Barcelona for Berlin, London and Austin in otherwise identical pages is a doorway pattern and gives a reader no new reason to trust the claim.

A European base can still be useful evidence. It explains time zones, languages, legal context and how remote delivery is organised. Present those facts directly. Do not disguise a lower-cost pitch as geography, and do not imply a local presence that does not exist. Correct hreflang links, self-canonicals and complete translations then connect the genuinely equivalent pages.

To check the language implementation, use the multilingual SEO checklist.

06 /Use llms.txt as a convenience, not a ranking promise

An llms.txt file can offer a compact, readable map of a site to tools that choose to consume it. It can also be useful for a person inspecting the site's main sources. Google now states plainly that it does not use llms.txt for Search visibility and that the file neither helps nor harms rankings. Creating one is therefore an optional publishing choice, not a GEO milestone.

The same caution applies to machine-readable summaries, special Markdown mirrors and blocks of hidden question variants. If they duplicate content, become stale or contradict the canonical page, they add another maintenance surface. Build them from the same source data when they serve a known consumer; otherwise spend the time improving the actual page.

Prioritise reliable responses, indexable HTML, correct status codes, accurate sitemap dates, accessible navigation and consistent information about the business. Check that production pages are not accidentally blocked. These controls support access and discovery without guaranteeing a citation.

07 /Measure visibility without inventing an AI ranking

Define the business outcome before collecting screenshots of answers. For a specialist service, useful measures include qualified organic enquiries, visits to a relevant service page, returning readers, branded searches and referrals from answer systems. A citation for a broad informational query may be interesting while producing no commercial value; an obscure guide that starts one good project can matter more.

Google's current Search Console guidance includes generative-search performance, while Bing Webmaster Tools' AI Performance reports citations, cited pages and sampled grounding queries. OpenAI adds a ChatGPT source parameter to referral links. These views measure different things and will change. Keep them separate rather than combining them into a made-up universal visibility score.

Review guides over a period long enough to observe impressions and queries. Improve missing explanations, update facts when primary sources change and create another article when it answers a distinct question. Record what changed, why and when, then review the results without assuming every change caused an increase.

For a repeatable measurement record, read how to measure AI citations and traffic.