ROOTCR HU

How does a small business get into ChatGPT and Google AI answers?

In short: Language models don't rank companies, they cite documents. For a model to mention your company, you need a page that on its own gives a self-contained, factual, verifiable answer to a specific question — and that robots can reach, read and understand in machine form too. This work is largely the same as what moves you forward in traditional search, just done more rigorously end to end. Nobody can guarantee it.

What has changed?

In the past, the search engine gave ten links and you chose. Today more and more questions receive a ready-made answer — from Google's AI overview through ChatGPT search to Perplexity — with two or three sources listed below the answer. Those few sources are the new first place.

Two abbreviations have spread for this: AEO (answer engine optimization) and GEO (generative engine optimization). They are about the same thing: how your content becomes a citable source.

What the model looks for — and what it doesn't

A language model doesn't look for “the best company”. It looks for a citable sentence for a question. That's why:

  • Doesn't help: text of the “many years of experience, reliable partner, quality service” type. This is not an answer to any question.
  • Helps: “A custom web store starts at $1,400 (€1,200), typically takes one or two weeks to complete, and the source code stays with the client.” This is a fact, it is citable, and it settles a question.

The six steps that actually matter

1. Have a document that can be cited

The most important step, and where most people fail. If only one document exists about your company — the service self-description — then you are not a source for any question. You need pages that answer one specific question each (“is it worth developing a custom web store?”, “how much does a website cost?”, “why is WooCommerce slow?”).

2. The beginning of the page gives the answer

At the beginning of every page and article there should be a 40–70 word, self-contained summary that answers the main question with subject–predicate structured facts. This is what an AI overview can lift out. (On this page too, this is the “In short” block.)

3. Question-form headings

The heading should be the searched question itself, and the first sentence should answer it directly. The model lifts content at paragraph level — not at page level.

4. Numbers with source and date

“The site is fast” — nothing. “TTFB 0,065 s, measured from our own server, August 14, 2026.” — this is citable. Models and people alike accept the sourced, dated number.

5. Machine readability

  • The content should be server-side rendered text, not appear only after JavaScript.
  • Robots should be given permission in robots.txt (GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended and their peers).
  • There should be structured data: who you are, what you do, at what price, told in machine form too. More on this separately: what is structured data.
  • It's worth serving an llms.txt and a per-page markdown mirror: a clean, markup-free version intended for the model. These few kilobytes are the cheapest way for it to receive your entire content in one request, without misunderstanding.

6. Unambiguous entity

Models think in entities. An unidentifiable company is a weak entity. What helps: a single canonical self-definition, literally the same everywhere (website, llms.txt, LinkedIn, email signature); a name and a face behind the company; own properties linking to each other; an imprint that identifies.

What can be measured?

This field is full of unfounded promises, so measurement is especially important:

What to measureHow
Whether the models reach your content at allServer logs: GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, bingbot hits weekly
Whether visitors come from AI interfacesReferring source: chatgpt.com, perplexity.ai, gemini.google.com
Whether they citeMonthly the same five questions to ChatGPT, Perplexity and Gemini, recorded

The third is the most important and the least automatable: ask the same questions monthly and see who appears as a source.

What nobody can guarantee

That you will appear. It doesn't depend on your page alone: the model decides, and the decision changes from month to month. Whoever promises a guarantee either doesn't understand it or doesn't say what they're selling.

What can be guaranteed: that the content is accessible, readable, factual, dated and interpretable in machine form too. Without these you certainly won't appear — with them you have a chance, and the same work also moves you forward in traditional search. That's the difference between an honest and a flashy offer.

Frequently asked questions

Is this the same as traditional search engine optimization?

Mostly yes, but the emphasis is elsewhere: the search engine ranks pages, the model cites paragraphs. That's why the self-contained summary, the question-form heading and the sourced number become more important.

Is blogging enough?

It's not quantity that counts. Eight to ten thoroughly written, number-backed, annually updated pieces are worth more than one weekly filler piece — models prefer the well-bounded, factual source.

How much does this cost?

If it's part of building the site, it's not a separate item: with me, structured data and machine readability are standard equipment on every custom website. Writing the content is what takes time — depending on whether the professional material comes from you.

Updated: August 23, 2026

Got a project that fits?

Describe it in a few sentences. I reply within one business day, tell you whether I can help, and give you a ballpark figure. If it is not a fit, I say that too.

Request a quote