Google says to skip llms.txt. Why do you still recommend it?

Correct, and we do not hide it: in 2026 Google stated explicitly that it does not use llms.txt and that there is no special schema for AI. The remaining argument is that the file is cheap to maintain and that consumers other than Google may read it. So we treat it as a low-cost measure, not a decisive factor — and we do not sell it as one. The real bot-access work happens in robots.txt, in rendering, and in content structure, which we recommend and your team applies.

Generative visibility

GEO — Generative Engine Optimization

I optimize you to be selected in AI-generated answers (ChatGPT, Perplexity, Google AI Overviews) — on both the retrieval and parametric channels.

Book a discovery call → Free SEO GEO AEO site check

What is GEO (Generative Engine Optimization)?

GEO is the discipline through which a brand gets selected and mentioned in the answers generated by AI engines such as ChatGPT, Perplexity or Google AI Overviews.

It works on two distinct channels: retrieval (the engine searches in real time and cites sources) and parametric (what the model already "knows" from its training data). The two call for different interventions — the first is solved through technical access and extractable content, the second through entity presence and mentions in sources the models have learned from.

All GEO, AEO and SEO definitions, in the glossary →

What's included

The retrieval channel

Content and structure that make you extractable in real time by bots that cite with attribution — the effect usually shows sooner than on the parametric channel.

The parametric channel

Brand and entity signals that influence what the model "knows" about you at the next retraining cycle.

AI bot access

Citation-first robots.txt (llms.txt optional): you allow search bots (OAI-SearchBot, PerplexityBot, Claude-SearchBot) instead of blocking them.

Brand entity

I position you as a clear, coherent entity, so models associate you with your commercial topics.

Why it matters

Search has moved from blue links to generated answers. GEO is the difference between being named or invisible in the client’s decision.

We work both channels (retrieval + parametric) — not just a one-off trick.

What a GEO engagement looks like in practice

A GEO engagement starts with a measurement, not an assumption: we run a set of questions relevant to your business across several AI engines, in clean sessions, and note whether the brand appears, how it is described and which sources are cited in your place. That produces a gap map: questions where you are absent, sources that replace you, outdated or wrong information about the brand.

The work itself splits across the two channels. On the retrieval channel we identify what stops a bot from picking you up (access in robots.txt, rendering, passage structure) and give your development team prioritized recommendations, while the pages that answer the question directly are content we write. On the parametric channel we work on brand coherence across profiles and third-party sources — the places a model builds its picture of you from. Every recommendation and every action is documented, so we know what produced what.

What we measure, and how

There is no "Search Console" for ChatGPT or Perplexity, so measurement is a protocol, not an automatic report. We combine several sources: repeated runs of the same prompt set (the same question can produce different answers and sources, so a single run proves nothing), AI-source traffic in GA4 and, for Google, the "Generative AI" report in Search Console.

We report separately what moved quickly (access and structure, where the effect can show once the bot re-crawls the page) and what builds slowly (the brand entity, which depends on model retraining). That way we don't confuse a session fluctuation with a real improvement.

What we do not promise

We do not guarantee positions and we do not guarantee citations: AI engines are probabilistic systems, and no provider controls what a model picks. We guarantee the agreed deliverables, delivered in full and on time. What you get is a verifiable diagnosis, documented work and a measurement that shows the trend, including when a change did not work.

Nor do we sell shortcuts: an llms.txt file or a particular schema type is not a proven citation lever. We recommend them where they cost little and clarify structure, but the core work remains bot access, answer quality and entity coherence.

Methodology

Our process for GEO — Generative Engine Optimization

Initial measurement

We run relevant questions across several AI engines, in clean sessions, and build a gap map: where you're absent and which sources replace you.

Retrieval channel

We give you the recommendations for bot access and passage structure (your team implements them), and we write the pages that answer the question directly.

Parametric channel

We work on brand coherence across the site, profiles, and third-party sources — where a model forms its picture of you.

Measurement & reporting

We re-run the questions and report separately what moved fast from what builds slowly.

Frequently asked questions

About GEO — Generative Engine Optimization

Classic SEO optimizes a page to rank in the list of results; GEO optimizes extractable passages so they get selected in a generated answer. What you measure changes (citations and mentions in answers, not only positions), so does the technical checklist (AI search bots allowed, correct rendering), and so does who describes you (consistent entity signals across the web, not only links). SEO stays the foundation: a page that is not indexed cannot be cited.

On the retrieval channel: passage-level structured content, extractable, plus allowed access for AI search bots (robots.txt) and correct rendering; llms.txt and schema markup are optional, as technical hygiene. On the parametric channel: consistent brand and entity signals across the web. No serious provider guarantees citation — the systems are probabilistic; the work increases your odds, and we measure it.

Retrieval = the model actively searches for content in real time and cites it with attribution; you control it through structure, authority, and crawlability, and the effect shows up sooner. Parametric = what the model internalized during training about you; it changes only at the next retraining cycle. The winning strategy aligns both channels.

GEO answers "am I picked as a source?" — getting into the set of sources an AI engine can cite, through bot access, entity coherence, and presence in sources models have learned. AEO answers "can a fragment be extracted from my page?" — it is about the shape of the content. You can be selected but uncitable, or citable but never selected, which is why we treat them separately. Worth knowing: the industry has no agreed definition separating these terms, and Google maintains both are still SEO. Full comparison, with what Google says →

On retrieval, changes can show up sooner, after content is structured correctly and bots have crawled it (there is no public measurement of the timing). On parametric, the effect comes more slowly — at the model’s next retraining cycle. That’s why we work on both simultaneously and report progress transparently.

Let's see where you stand.

A 30-minute discovery call. No pitch — I'll tell you directly if and how I can help with GEO — Generative Engine Optimization.

Book a discovery call → 0727 542 757 WhatsApp