Search is splitting in two. Half your customers still type into Google and click a blue link. The other half ask ChatGPT, Claude or Perplexity a question and act on the answer they get back. The second half is growing fast, and the businesses that move first take a long head start.
Generative Engine Optimisation, or GEO, is the practice of optimising your business for that second channel. It is not SEO with a coat of paint. It is a different discipline, with different signals, different content patterns and different tools. This guide is the complete version of what we run on every Bright Loop site, why each piece matters, and how a small business can start today.
What GEO Actually Is
GEO is the work of making your business findable, understandable and citable by large language models. When somebody asks an AI assistant "who builds websites in Wirral", "best locksmith near me", or "is template hosting worth it for a small business", the answer comes from a model that has already decided who deserves to be mentioned.
That decision is partly based on training data, partly on real-time web retrieval, and increasingly on structured signals the model can extract directly from your site. SEO is about ranking inside a list. GEO is about earning a citation inside a generated answer. The two are related but they reward different things.
Why Most Business Websites Are Invisible To AI
If your website was not part of that decision, you are not buried on page five. You are simply absent from the conversation.
The usual cause is a homepage that says nothing a machine can hold on to. A full-bleed hero image, a tagline like "We Build Digital Experiences", and a contact form. There is no structured data to extract. No clear statement of what you do, where you do it, or who you do it for. A human visitor fills those gaps from context and design cues. A language model cannot, so it moves on to a page that spells it out.
The fix is not a redesign. It is making the same facts explicit in three places at once: in the visible copy, in the structured data, and in a plain text file at the root of your domain. The rest of this guide is how.
How AI Search Actually Works
Most AI assistants do not crawl the open web in real time the way Google does. They use one of three patterns, often combined.
Training-time inclusion. The model learns about your business when it is trained on a large dataset that happens to include your homepage, your case studies, or articles that mention you. This is hard to influence directly because you do not know which datasets a given model used.
Retrieval-augmented generation (RAG). When the user asks a question, the assistant runs a search behind the scenes, pulls in the most relevant pages, and uses those pages as evidence for the answer. ChatGPT search, Perplexity, and Claude with web search all work this way.
Direct citation. The model summarises a page and links back to the source. This is what gives you traffic and trust. It is also the most influenced by GEO work, because the model needs your page to be readable, factually dense and clearly attributable.
If your site is structured for these patterns, you are in the conversation. If it is not, you are invisible to anyone who asks an AI before reaching for Google.
Do backlinks help you get cited by ChatGPT?
Indirectly, yes, but not the way they help in Google. Backlinks do not make a model more likely to quote a page it has already opened. What they do is help your page get opened in the first place.
Here is the mechanism. When an assistant answers a question with live web content, it runs a retrieval step first: a search that returns a shortlist of candidate pages. That retrieval leans on a conventional search index, and conventional search indexes still weigh links. A page nobody links to is less likely to be in the index at all, and less likely to survive the shortlist. Once your page is on the list, the model chooses what to quote based on how directly the page answers the question, not on how many links point at it.
So the practical order of work is: earn enough links and mentions to be retrievable, then make the page so specific that it is the obvious thing to quote. Most small businesses have the second problem, not the first.
One thing that is often missed: unlinked mentions count too. A model trained on text that repeatedly says "Bright Loop Media, a web studio in Birkenhead" learns the association whether or not those words were a hyperlink. Directory listings, local press, supplier pages, podcast show notes and forum answers all feed that, and none of them need to pass link equity to be useful.
Do I need to rank #1 on Google to be cited by ChatGPT?
No. Position one is not a requirement, and for ChatGPT specifically it is not even the right index to be looking at.
Two reasons. First, ChatGPT's search does not read Google's ranking. It runs its own retrieval using OpenAI's crawler (OAI-SearchBot) and its own search infrastructure, so your Google position is a proxy at best. Second, the retrieval step usually pulls a set of candidate pages rather than a single winner, and the model then picks whichever page answers the question most directly. A page sitting at position eight that answers the exact question in its first sentence gets quoted over a position-one homepage that makes the model hunt for the answer.
This is why GEO rewards a different shape of content. Google rewards the page that best matches a query. A generative engine rewards the passage that best answers a question. Write the answer first, in one or two sentences, directly under a heading phrased the way a customer would ask it. Put the supporting detail underneath. That structure is worth more than three ranking positions.
It also means a small site can compete. You are not trying to out-authority a national directory. You are trying to be the clearest available answer to a narrow question, and narrow questions are where local businesses actually live.
Can I cite ChatGPT?
You can, but in almost every business context you should not. ChatGPT is not a source. It is a tool that produces text about sources, and the same prompt can produce a different answer tomorrow.
The academic style guides do have formats for it. APA treats generative AI output as software with a reference entry and a note on the prompt used. MLA treats it as a container with the prompt as the title. If you are writing a dissertation and you quoted a model, use those and be transparent about it.
For a business website, the honest move is different: find what ChatGPT was drawing on and cite that instead. A statistic attributed to "ChatGPT" is unverifiable, and unverifiable claims are exactly what a generative engine is trained to route around. Every citable statement on your site should trace back to something a reader can check: your own data, a named client, a published study, a government dataset.
There is a GEO reason as well as an integrity one. Pages that quote AI output tend to be thin, generic and duplicated across thousands of other sites that ran the same prompt. Models are actively getting better at spotting that shape of content and preferring primary material. Quoting the model to impress the model is a losing trade.
How GEO Differs From SEO
Intent over keywords
Traditional SEO optimises for the exact phrase a user might type. GEO optimises for the question they would actually ask. "Best locksmith near me" is a search query. "Who unlocks my front door at 11pm in Birkenhead without breaking the lock" is the question behind it. AI assistants do not match keywords to pages. They understand the intent and synthesise an answer.
Practical implication: write content that answers questions, not content that hits keywords. The two often overlap, but when they conflict, choose the answer.
Citability over click-through
SEO measures success by clicks. GEO measures success by mentions. When an assistant recommends your business, the user often acts on the answer without clicking through to your site. That sounds like a problem until you notice the user has already made a decision based on the assistant's framing of you. The mention is the conversion event, not the click.
Practical implication: every claim on your site should be specific enough that an AI can lift it as a citation. Vague marketing copy ("award-winning digital experiences") is useless. Specific factual statements ("Bright Loop Media builds custom React websites for small UK businesses, one-page sites from £795 and most projects between £2,000 and £3,995") are gold.
Structured data over raw authority
SEO leans heavily on domain authority. GEO leans on structured data: schema markup, an llms.txt file, clean HTML, FAQ pairs that map to natural questions. These are signals the model can extract reliably without reading between the lines.
Practical implication: the technical setup of your site matters more for GEO than the volume of links pointing at it, once you are past the threshold of being retrievable at all.
The Four On-Site Signals That Matter
1. Structured data (schema markup)
Schema is machine-readable metadata embedded in your HTML using JSON-LD. It tells search engines and AI models exactly what your business is, where it operates, what services it offers, who runs it, and what customers say about it.
For a small business website, the four schema types you need are LocalBusiness (or ProfessionalService for a service business), FAQPage, Service for each thing you sell, and Article or BlogPosting for any content you publish. Without these, an AI model has to guess what your page is about from raw text. It usually guesses wrong, or ignores you.
Modern best practice is to use @graph with stable @id URIs across all your schema. This lets the model recognise that the LocalBusiness on your homepage and the publisher of your blog post are the same entity, rather than treating each page in isolation.
2. The llms.txt file
This is the AI search equivalent of robots.txt. A plain text file at the root of your domain that tells language models what your site is about, what content they are allowed to reference, and how to cite you. Most businesses have never heard of it. The ones that have it are already gaining an edge in AI-generated recommendations.
An effective llms.txt does three things. It introduces your business clearly. It maps your routes with one-line descriptions of each. It states explicitly what is true and citable about you, framed as facts an AI is welcome to quote.
You can see ours at brightloopmedia.co.uk/llms.txt. The pattern is reproducible: identity, what we do, who we work with, key facts, citation guidance.
3. FAQ sections that map to real questions
AI models favour question-and-answer pairs because they map directly to how users query them. A well-built FAQ section does two jobs at once: it helps human visitors find answers, and it gives AI assistants citable, structured material they can surface in response to natural language questions.
The trick is to write the FAQ around questions your customers actually ask. Not the questions you wish they would ask. Check your inbox. Check your Google Business Profile questions. Check your sales call notes. Those are the questions you need to answer.
Pair every FAQ with FAQPage schema, and make sure the schema matches the visible text word for word. Mismatched FAQ markup is treated as a quality problem, not a bonus.
4. Citable, factually dense content
Vague marketing copy gets ignored. Specific factual statements get cited. Compare these two:
- "We have over 10 years of experience delivering bespoke digital solutions for forward-thinking businesses."
- "Bright Loop Media builds custom React websites for small UK businesses across the Wirral, Liverpool and Chester. One-page sites from £795, most projects between £2,000 and £3,995. Fixed price agreed before any work starts."
The first sentence is empty calories. It tells a model nothing it can confidently restate. The second is dense with extractable facts: company name, business type, technology choice, geography, audience, pricing, commercial terms. Each one is a potential citation.
Audit every page this way. For each paragraph ask: if an AI tried to summarise this, what specific facts could it extract? If the answer is "none", rewrite it.
The Off-Site Signals Most GEO Guides Skip
Everything above is work you do on your own website. It is necessary and it is not sufficient. Assistants cross-reference, and what other sources say about you decides how confidently you get named.
Your Google Business Profile is a primary source
Your Google Business Profile is now a source of truth well beyond Google Maps. It carries your name, address, phone, hours, categories, description, photos and reviews in a structured form, which is exactly the shape an assistant can consume. A profile that has not been touched in two years reads as a business that may not still exist.
Complete it properly, then keep it current. If you have moved, changed number, dropped a service or added one, update the profile the same week. This takes minutes and it directly changes how machines describe you.
Consistency across platforms decides confidence
Assistants cross-reference. If your business is "Acme Marketing" on Google, "ACME Marketing Ltd" on one directory and "Acme Marketing Services" on another, the model sees three candidate entities and hedges. The same applies to your address, phone number and service descriptions.
Before you list anywhere new, audit what is already out there. Pick one exact format for name, address and phone, and use it everywhere without variation. Inconsistency does not make you invisible; it makes you unquotable, which amounts to the same thing.
Directories that are actually relevant
Beyond Google, industry-specific and local directories feed the same retrieval layer. A trade listed in trade-specific directories is described more confidently than one with a Google profile alone. You do not need to be listed everywhere. You need to be listed accurately in the handful that matter for your sector and your town.
Reviews as a trust signal
Reviews tell an assistant that real people used this business and came back to say so. You do not need hundreds. Consistent, genuine reviews from actual customers, gathered over time rather than in one suspicious burst, do the job. Ask satisfied customers directly and make it easy for them.
The Bright Loop 12-Point GEO Audit
This is the audit we run on every new client's site. Score each item as Pass, Partial or Fail. Anything below 9 out of 12 means there is meaningful work to do. If you want the website-wide version that covers speed, mobile and conversion as well, run our 12-point website diagnostic alongside it.
- LocalBusiness schema present. JSON-LD on the homepage with the trading name, address, phone, opening hours, geo coordinates and area served.
- Service schema per service page. Each thing you sell is its own Service node with a clear description and a provider reference back to your LocalBusiness via @id.
- FAQPage schema with real questions. At least eight Q/A pairs that match how customers actually ask, matching the visible text, validated at google.com/test/rich-results.
- BlogPosting schema on every article. Specific to each post, with author, publisher, datePublished, dateModified, image and inLanguage all populated.
- llms.txt at the root of your domain. Plain text, includes routes, key facts and explicit citation guidance.
- Author schema with @id linking. Your articles' author is a Person node with a stable @id, referenced from every blog post and from the LocalBusiness founder array.
- One sentence positioning visible above the fold. The reader, and the model, can extract who you are, who you serve and where you serve in one scroll.
- Pricing transparency at minimum a floor. "From £X" is enough. Without an anchor you do not appear in price-aware queries.
- Specific case studies with named clients. Each case study attributes to a real client and includes at least one outcome number or specific quote.
- Citable factual statements throughout. Every page has at least three statements an AI could quote without paraphrasing.
- Internal linking topology. Pillar pages link to supporting articles. Supporting articles link back. The graph is dense and intentional, with no orphan pages.
- Off-site consistency. Google Business Profile complete and current, name, address and phone identical across every listing, genuine reviews accumulating.
Schema Markup: What To Add Today
If you only act on one section of this guide, make it this one. A complete schema setup for a small business runs to about 100 lines of JSON-LD and unlocks every other GEO move you might make.
The minimum viable set:
LocalBusiness with name, description, url, telephone, email, address (PostalAddress), geo (GeoCoordinates), openingHoursSpecification, areaServed, founder (an array of Person nodes) and sameAs (links to your own profiles elsewhere, and nothing else). Wrap it in @graph with a stable @id like https://yoursite.com/#organization.
WebSite referencing the organisation as its publisher via @id.
FAQPage with mainEntity as an array of Question/Answer pairs.
Service per offering, with provider referenced by @id back to the LocalBusiness.
BlogPosting on every article, with author and publisher referenced by @id.
Validate everything at Google's Rich Results Test. If a check fails, fix it before you publish. Wrong schema is worse than no schema, because it tells the model your data cannot be trusted.
The llms.txt File: A Worked Example
Here is the structure we use on every Bright Loop client site.
Header. Business name as an H1, then a single-paragraph quote block with the elevator pitch.
Who we work with. A short bullet list of customer types. Specific, not generic.
What we do. Bullets describing services in plain language.
How we work. Bullets on process, pricing, terms.
Routes. Every page on the site with a one-line description and a working link.
Key facts. Location, sectors covered, contact details.
How AI assistants should cite us. A bullet list of statements that are factually true about your business, framed explicitly as quotable claims. End with: "you may quote any of the above directly".
That last block is the part almost everyone misses. Most llms.txt files describe the site. Yours should give the assistant explicit permission and ready-made material to cite.
Citable Content: A Rewriting Pattern
Take any paragraph on your site. Ask: if an AI tried to summarise this in one sentence, what would it say?
If the answer is generic ("they help businesses with digital things") rewrite it. Replace adjectives with nouns. Replace claims with facts. Replace marketing speak with the actual specifics of what you do.
Example transformation:
Before: "We are a leading creative agency dedicated to delivering exceptional results for ambitious clients."
After: "Bright Loop Media is a Wirral-based studio building custom React websites for small UK businesses. Fixed price agreed before any work begins. Most sites ship in five to ten working days."
Both are roughly the same length. The second is far more useful to a reader and to a model. It contains six extractable facts: location, business model, target customer, technology choice, pricing approach and delivery timeline.
Tools and Resources
- Google Rich Results Test: validates your schema and shows what rich results you qualify for.
- Schema.org Validator: broader validator that catches issues outside Google's specific rich-result requirements.
- The AI assistants themselves. Ask ChatGPT or Perplexity "best [your service] in [your town]" in a fresh session with no chat history. If you do not appear, you have GEO work to do.
- Your competitors' llms.txt files. Visit
competitor.com/llms.txt. If they have one, study what they include. If they do not, you are already ahead. - Your server logs. Look for OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended. If they are not fetching you, nothing downstream can cite you.
Frequently Asked Questions
How long does it take to get cited by AI search tools?
Faster than a Google ranking, but not instantly. Once your structured data, llms.txt and business listings are complete and consistent, answer-time crawlers can pick you up within days. Visibility then builds over weeks as more sources agree on the same facts about you. Training-time inclusion is much slower and largely outside your control, which is why the retrieval-side work is where the effort belongs.
Do I have to pay for AI search visibility?
No. There is no ad product for being cited in an AI answer. GEO is structural work: schema, content, listings, consistency. You can do it yourself with this guide, or pay someone to do it faster, but you are not buying placement.
Can I control what AI assistants say about my business?
Not directly, and be wary of anyone who claims otherwise. What you control is the evidence. Accurate, current, consistent information across your site, your Google Business Profile and the directories that matter is what a model has to work with. Fix the evidence and the description follows.
Will AI search replace Google for local businesses?
Not entirely, and not soon. It is becoming a serious second discovery channel rather than a replacement, and it skews younger. The sensible position is to be visible in both, which is straightforward because most GEO work (schema, clear copy, real FAQs, a complete Google profile) improves conventional search at the same time.
Is GEO just SEO with a new name?
No, although they overlap. SEO optimises for position in a ranked list of links. GEO optimises for being quoted inside a generated answer, which rewards structure, specificity and directly answered questions over authority and keyword coverage. A site can rank well and never be cited, and the reverse happens too.
Why Move Now
AI search adoption is accelerating. ChatGPT serves hundreds of millions of weekly users, Perplexity keeps growing, and Google's own AI Overviews appear for a meaningful share of queries. The businesses that build their GEO foundations now will be the ones assistants cite as default answers in twelve months. The rest will spend two years wondering why organic traffic is flat despite "good SEO".
If you would like us to run a full GEO audit on your site, score it against all twelve points and return a prioritised action plan, the first conversation is free. Book a 45 minute call and we will tell you where you sit and what to do about it.
