AI SEO: The 2026 Guide for Small Businesses

Half your customers still Google. The other half are increasingly asking ChatGPT, Claude, Perplexity or Gemini. A growing share never see a classic results page at all, they read a single AI-generated answer and act on it.

If your website is not optimised to be that answer, you are invisible to roughly a third of your market. This guide explains, in practical terms, how a small business in Watford, Hertfordshire or anywhere in the UK can fix that.

This guide is the strategy. For the build detail, how the HTML, headings and discovery directives actually need to be put together, see hand-coded site AI visibility. If you would rather we just did it, that is our AI SEO service.

What AI SEO actually is

AI SEO, also called GEO (Generative Engine Optimisation), is the practice of making your website easy for large language models to:

1. Crawl: so they can read your content at all.
2. Understand: so they grasp what you sell, where, to whom and with what proof.
3. Cite: so when a user asks “best accountant in St Albans” they get your name, your URL and your phone number.

The good news: about 80% of AI SEO is exactly what great classic SEO always has been. The remaining 20% is where most businesses are currently missing out.

The AI SEO foundations (that are also classic SEO)

• Fast, accessible, crawlable site. If Googlebot struggles, so do LLM crawlers.
• Clean semantic HTML. Proper <h1>, <h2>, <article>, <nav>, not a soup of <div>s.
• Concise paragraphs with a single clear fact each. LLMs love pages where one paragraph = one citeable answer.
• Internal linking with descriptive anchor text. “Our CRM development services” beats “click here”.
• Honest facts. Prices, addresses, phone numbers, years in business, specific and consistent site-wide.

The AI-specific 20%

1. Structured data everywhere it makes sense

JSON-LD Organization, LocalBusiness, Service, FAQPage, BreadcrumbList, Article, Product. Every LLM citation we have traced back to a specific technical pattern includes rich, valid structured data on the source page.

2. An /llms.txt file

llms.txt is an emerging standard (proposed 2024, now widely adopted) for summarising your site for large language models. A short, machine-readable index of your most important URLs and what they cover. It is the robots.txt of the LLM era.

3. Entity-linked content

Instead of writing “we serve the surrounding area”, write “we serve Watford, St Albans, Hemel Hempstead and the WD, AL and HP postcodes”. LLMs build their answer from the entities named on the page.

4. Genuinely useful FAQs

Every key service page should have 4-8 frequently asked questions with direct, one-paragraph answers. These are what LLMs most commonly lift into their responses.

5. Freshness signals

Update key pages at least twice a year and include a visible “last updated” date. LLMs that cite via retrieval (Perplexity, Bing Copilot, Google AI Overviews) weight freshness heavily.

Making your site agent-ready: what we did to ours

There is a newer layer on top of all this, and in July 2026 we put our own site through it, so this is first-hand rather than theory.

Here is what is changing. AI agents, not just chatbots but tools that browse and act on someone’s behalf, are starting to read websites differently from both people and classic crawlers. Cloudflare now runs an “agent-readiness” scan that scores how easy your site is for those agents to use. We ran it on stuffedmonkeywebdesign.com and scored 21 out of 100. For a company that sells AI-search services, that stung. Here is what actually worked, and what we left alone.

Content signals in robots.txt. There is a new Content-Signal line you can add to robots.txt (see contentsignals.org) that states, in machine-readable terms, whether you are happy for your content to be used for search, for AI answers and for model training. A lot of sites are quietly hedging on this. We set all three to yes, because being quoted by ChatGPT and cited in an AI Overview is exactly what we want. If people finding you is good for business, do the same. If you sell something you would rather not have scraped wholesale, that is a fair call too, but say so plainly instead of leaving it vague.

An llms.txt file, kept current. We covered llms.txt above. Our mistake was writing it once and forgetting about it. Ours had drifted months out of date, missing whole pages and listing an old title. An llms.txt that is wrong is worse than none at all. Treat it like a sitemap for machines and update it when the site changes.

A Markdown version of every page. HTML is noisy for a machine: navigation, styling and scripts wrapped around the few paragraphs that actually matter. So we now publish a clean Markdown copy of every page and tell agents it is there, with a link tag and an HTTP header. An agent that just wants the facts can grab the Markdown and skip the rest. On a static site this is a build step, not a rebuild.

What we deliberately skipped. The same scan rewards things that only make sense if you are running an app: a live agent-facing API, login for agents, machine-readable checkout and payment. We have no login and no checkout, so building hollow versions of those would have been dishonest, and would have scored us points for things we cannot actually do. We left them out. A brochure site should do brochure-site things well, not pretend to be a piece of software.

None of this is exotic. It is a line in robots.txt, a text file, a build step and a couple of HTTP headers: an afternoon’s work on a site that was already built properly. The goal is the same as it has always been in SEO. Make it easy for the machine to understand you, then get out of its way.

What will NOT work

• Keyword stuffing. LLMs are better than Google ever was at detecting it.
• Hidden text or schema that does not match what is on the page, you will get quietly demoted.
• “AI-generated content at scale”. Generic filler without original insight gets ignored by the very models that produced it.

Where to start tomorrow morning

1. Run a free Lighthouse audit and fix any Core Web Vitals in the red.
2. Add Organization and LocalBusiness JSON-LD to your homepage.
3. Pick your three most important service pages and add 5 FAQs each with FAQPage schema.
4. Create an /llms.txt file listing your 10-20 most important URLs.
5. Ask ChatGPT, Claude and Perplexity “best [your service] in [your town]”, and see who gets cited. That is your competitive set.

If any of this sounds daunting, get in touch, we offer a fixed-fee AI SEO audit that covers all of the above and gives you a prioritised roadmap.