What GEO means for a small business site in 2026

Generative engine optimization — GEO — is what people started calling the work of being readable and quotable by systems that answer a question instead of handing over ten links. ChatGPT, Perplexity, Google's AI Overviews and Bing Copilot all do the same three things: they fetch pages, they lift a passage, and they attribute it. GEO is the work of making all three easy. There is no separate submission form, no GEO account, and nothing to buy.

That is the short version, and it is worth saying plainly because the acronym is doing a lot of marketing work right now. The practical question for a five-page plumbing site or a twelve-page accountancy is narrower and more useful: when a machine reads my page, can it find a sentence it would be willing to repeat, and will it know the sentence came from me?

What actually changed

Two things, and only two.

The first is the shape of the reader. A classic search crawler indexed your page so a human could decide whether to click. An answer engine reads your page so it can write a paragraph. It is not scanning for keyword density; it is scanning for a claim it can stand behind. A page that circles the topic for four paragraphs before saying anything specific gives it nothing to lift.

The second is where the attribution goes. In a ten-blue-links world, being the source meant being the destination. In an answer, being the source can mean one citation chip under a paragraph the reader may never expand. That is a real loss of traffic and pretending otherwise helps nobody. It is also why the fundamentals matter more, not less: the citation you do get is the entire exposure.

The part most small sites get wrong first

Before any of the writing advice, check whether the machines can reach you at all.

Cloudflare, which sits in front of a very large share of small business sites, ships a setting that blocks AI crawlers, and on new zones it has defaulted to on. A site can have perfect content, a clean sitemap and a permissive robots.txt, and still be invisible to every answer engine, because the block happens at the edge — above robots.txt, before your server is ever asked.

This is not theoretical. It is the single most common reason a site does everything else right and appears in nothing. Test it from a terminal rather than trusting a dashboard:

curl -s -o /dev/null -w "%{http_code}\n" \
  -A "GPTBot/1.2" https://example.com/

A 200 means the door is open. A 403 means you have been doing GEO with the lights off. Repeat it for ClaudeBot, PerplexityBot and OAI-SearchBot; they are blocked independently, and one passing does not prove the others do.

Four things that are worth doing

Answer in the first two sentences. Put the claim, with the number, at the top of the page — then explain. This is not a style preference. Answer engines quote at the passage level, and the passage they reach for first is the one that resolves the question. A page that opens with "there are many factors to consider" has excluded itself.

Make one page per question a real person asks. Not per keyword — per question. "How much does a boiler service cost in Leeds" is a page. "Boiler services" is a category. The first can be quoted; the second can only be ranked.

Keep your structured data honest. If you mark up an FAQ, the questions and answers in the markup must be the ones visible on the page. Invented schema questions used to be a cheap trick; now they are a fast way to have your markup ignored entirely, and the engines that cite you are checking the rendered page, not just the JSON-LD.

Give every page a real link from somewhere. A sitemap is a verification list, not a discovery channel. Pages that exist only in the sitemap and are linked from nowhere sit in Search Console as "URL is unknown to Google" for months. If you publish an article, link to it from the page it belongs under.

What to ignore

Anything sold as a shortcut into an answer engine. There is no index to pay into. The services offering "AI citation placement" are selling either ordinary PR or link schemes, and link schemes are an old problem with a new label.

Ignore, too, the advice to rewrite your whole site in a "conversational tone for AI." The engines are reading for substance, not register. A specific sentence in a dry tone is quoted; a chatty page with nothing to quote is not.

How you can tell whether it is working

Badly, if you expect a clean number — and it is better to say so than to invent one. Referral traffic from chat interfaces is systematically under-reported: many arrive with no referrer at all, and none of them appear in Search Console, which only reports Google Search.

The measurements that do hold up are indirect but real. Your server logs show which AI user agents fetched which URLs and how often — that is a fact, not an estimate. Search Console still reports impressions and average position for the queries where AI Overviews appear, and a page that loses clicks while holding impressions is telling you something specific about where the answer is being read. And you can simply ask the engines your own questions on a schedule and record whether you are named.

None of those is a dashboard number. Together they are enough to know whether the work is landing, which is the only thing the number was for.

The honest summary

GEO in 2026 is not a new discipline. It is the same three questions it has always been — can a machine fetch this page, can it find a sentence worth repeating, and does it know who said it — asked by a reader with less patience and a much shorter output. The sites doing well are not the ones that found a trick. They are the ones that let the crawlers in, said something specific in the first paragraph, and kept their markup matching their page.