Key takeaways
- Search and answer engines should repeat the same facts. GEO is extractability of one public truth, not a filename or a schema trick.
- Put the same claims in the title, H1, body, and schema. Do not invent a second story in llms.txt, structured data, or social bios.
- CLICK.BLUE will put one public truth on the pages — the site, the copy, and SEO — not a second company written for machines.
The title says you design websites. The H1 says you are a growth platform. The schema lists a service the visitor cannot find. A chat tool describes a company that is not there.
That is how SEO and GEO for small businesses usually fail. Search and answer engines should repeat the same facts. Generative engine optimization is not a new ranking lever. It is whether a cautious machine can extract the same entity a person just read — the name, the offer, the proof, the next step — without reconstructing a cleaner company from slogans, schema, or a file at the root.
One entity. One claim set. If they hear two, you taught them a fiction.
Two readers, one company
SEO is still the retrieval problem. A crawler has to find the page. An index has to understand what the URL is about. A ranking system has to decide whether that URL is useful for a query. Titles, headings, visible copy, internal links, speed, and demand still matter. None of that is new, and none of it is a promise that a page will rank.
GEO — generative engine optimization — is the extractability problem. A chat tool or a generated search answer may name a company, a service, and a fact. It may shrug. It may invent a better version of you from a bio that does not match the homepage. The useful question is not "how do we rank in ChatGPT." The useful question is: if a cautious model quotes us, can it do so from a sentence we already published and will stand behind?
Answer engine optimization and entity SEO are the same discipline under different names: who you are, what you sell, what you do not sell, how someone starts work. The acronyms multiply. The work does not.
| Reader | Job | What it needs |
|---|---|---|
| A person | Decide whether you are real and relevant | A sentence they can finish, proof they can check, a next step |
| A search engine | Retrieve the right URL and understand it | One canonical page, crawlable HTML, copy that matches the title |
| An answer engine | Repeat a short fact without inventing one | An extractable entity: name, offer, proof, already visible |
Three readers. One page. One story.
If the homepage says you design and build websites, the title should not say you are a growth platform. The H1 should not say you are an AI company. The schema should not add a service the visitor cannot find. The llms.txt file should not invent a metric the HTML will not admit. That is GEO when the phrase means anything: extractability of one public truth, not a metadata trick.
Maintain an entity fact sheet
Create one governed record for facts that recur across the site, Business Profile, social profiles, structured data, proposals, directories, and llms.txt.
Include:
- canonical organisation name and spelling;
- founding date only if documented;
- locations or service area;
- primary offer family and plain-language descriptions;
- named products or productised services;
- current contact and next step;
- public proof with source and permission;
- author and editorial disclosures;
- canonical URLs for each major fact;
- owner and review date.
This is not a keyword sheet. It is the public entity’s change-controlled data.
Then map every fact to visible HTML. Structured data can classify it. llms.txt can point to it. Social profiles can compress it. None should introduce a claim without a canonical public source.
Google’s 2026 generative-AI resource says valuable, unique, non-commodity content matters and that SEO foundations remain relevant. Search Console has also begun testing dedicated generative-AI visibility reports for some sites. Use those reports when available, but do not confuse impressions in an AI feature with authority, qualified traffic, or revenue.
Measure search and generative visibility alongside engaged visits, internal movement, enquiries, references, and assisted conversion. The fact sheet protects consistency; the article and service pages provide the substance worth finding.
Read Google Search Central — Optimising for generative AI and Google Search Central — Generative AI performance reports.
Facts that survive four places
A public truth is a claim you will repeat in every place a stranger, a crawler, or a model might look.
For a small business the list is short:
- The legal or trading name, written the same way.
- What you sell, in nouns a buyer uses.
- Who it is for, if that constraint is real.
- How someone starts — a form, a call, a booking.
- The proof you can actually show.
- The founding year, the operating model, the contact route, if you publish them at all.
Write that list once. Then put the same claims in the title, the H1, the body, and the schema. If a claim cannot survive that repetition, it is not ready to publish. It is a slogan.
Search snippets often come from visible page text, not the meta description you wrote. An answer engine that cites you will lift a short, supported sentence from HTML, or it will skip you. Schema that introduces a claim the visitor cannot read is a second story. A second story is how entities get blurred.
Title. Name the same job the page does. Do not bait a query the body will not answer.
H1. Same claim as the title, in language a buyer would say out loud.
Body. The first screen makes the fact literal. Prefer concrete nouns: website, offer, form, launch. A person should know what you are and what to do next before they scroll.
Schema. Clarify a visible fact. Do not introduce a service, a review, a price, or a proof the visitor cannot read.
Those four places are one sentence, restated.
Slogans cannot be quoted
A model will not reconstruct your business from "innovative solutions" or "tailored excellence." Those phrases contain no entity. They do not name a service. They do not name a next step. They are easy to write and impossible to quote without making something up.
Extractability looks like ordinary sentences:
- We design and build brands, websites, and digital products.
- The scoped site is designed, written, and launched.
- Project intake is a form.
Those sentences can be lifted. They can be checked against the page. They can appear in a title without becoming a different company. That is entity SEO for a small business: make the nouns stable, visible, and boring enough to repeat.
What extractability is not: a JSON-LD block that lists services the page never describes. A filename at the root that "adds GEO." A social bio that claims a category the site does not support. A promise that any of those will rank you or get you cited.
The cheaper story usually lives in metadata
The second story is usually cheaper than the first. That is why it appears.
A plugin dumps every possible type. LocalBusiness, ProfessionalService, Product, FAQ, Review — whether or not the page shows those things. The visitor sees a five-page offer site. The graph sees a marketplace. An answer engine that trusts the graph over the prose will describe a business that is not there.
Someone hears that agents read a Markdown file at the root and writes a better about-page than the homepage. They add a metric. They add a service. They add a tone the brand does not use. The file is public. A person can open it. If it disagrees with the site, both readers learn you are sloppy. What llms.txt is — and what it is not is the longer version: a README, not a ranking lever. Google Search does not use the file. Write one after the pages exist.
The LinkedIn line says "AI-powered growth." The X bio says "full-service agency." The Google Business Profile, if you have one, says something else again. External surfaces corroborate an entity only when they repeat it. A contradiction across profiles is entity ambiguity. It is not distribution.
A listing that uses a different name or category is a second company with your URL attached. An empty accurate listing is better than a rich invented one. A thin URL stuffed with synonyms is the same pattern. People bounce. Models that quote it quote mush.
Markup cannot mint a citation
An answer engine may quote a short, supported sentence from visible HTML. It may ignore you. Schema and llms.txt may repeat that sentence. They cannot mint a citation.
Google Search does not use new AI text files or special markup to rank pages or to appear in its generative features. It is fine to maintain llms.txt for other systems. Doing so will not help or harm Google visibility. If the homepage is still a slogan, the wrong place to spend Monday is a README at the root. The right place is the pages.
Markup that matches the page can help a system understand type and entity. Markup that invents reviews, prices, or services is a trust problem.
The order of work:
1. Decide what the business actually sells. 2. Put that on public pages a stranger can finish. 3. Make those pages crawlable, fast enough, and internally linked. 4. Repeat the same claims in the title, H1, body, and schema. 5. Align the bios and listings you already have. 6. If you want a note for agents, summarize the pages you already shipped.
Skip step six if you have not finished step two.
Align what you already published
You do not need a knowledge-graph program. You need a short list of facts that survive contact with a skeptical buyer.
One spelling of the name. What you sell, in the buyer's nouns. What happens when someone says yes. Only the proof you can show. For CLICK.BLUE that is the public record: founded in 2014, more than twelve years, more than fifty businesses, more than a hundred websites. Those numbers appear on the site. If you do not have a number, do not mint one. A named process is better than a fake metric.
What you do not do is part of the entity. When the offer changes, change the title, the H1, the body, and the schema in the same week. Monitoring a chat tool does not itself improve visibility. It tells you where the public truth is thin.
Once the facts are governed, a website and search programme can publish and maintain them across the relevant pages. Start with the sentence, then the four places, then the bios.
CLICK.BLUE will put one public truth on the pages and keep it there. When the gap is implementation, the estimator translates the current public system into a scoped website or search programme.
