A friend of mine runs a small skincare shop online. Business had been steady for two years, then quietly dipped for three months straight — no obvious reason, no algorithm update she could point to, nothing wrong with the site. Then a customer messaged her: “I almost bought from [a competitor] instead — I asked ChatGPT which brand to get and it mentioned them first.” She hadn’t lost a search ranking. She’d lost something she didn’t even know she was being scored on.

That’s the situation this piece is written for: not “should I use AI for marketing,” but “customers are already asking AI which business to pick, and mine isn’t coming up.”

This isn’t SEO, and treating it like SEO wastes your time

Traditional SEO gets you a slot in a list of ten blue links, ranked by position. What’s happening now is different: a customer asks a chatbot a direct question — “best [your category] in [your city],” “which [product] is actually worth it” — and the model picks a small handful of businesses to name, out of everyone it could have picked. There’s no ranking to climb. You’re either mentioned or you’re invisible, and there’s no page two.

The industry has started calling this GEO — Generative Engine Optimization. It’s the discipline of structuring your content and online presence so language models retrieve, evaluate, and reference your brand when they generate an answer, rather than optimizing for a ranked list on Google. It’s also less mature than SEO, which is good news if you’re small: most mid-market businesses haven’t started doing this seriously yet. The gap between “everyone does this” and “almost nobody does this” is exactly where a small business can outrun a much bigger competitor.

Why this isn’t a niche worry anymore

The scale of the shift is the part most small business owners haven’t clocked yet. Industry tracking in 2026 puts the share of US consumers who now use AI at the product-discovery stage at around 35%, against roughly 14% who still start with a traditional search for the same job — meaning for a growing share of your potential customers, the shortlist is decided before they ever open a search bar or see your website. Perplexity alone is now handling on the order of 780 million queries a month, and that’s just one of several AI answer surfaces sitting between a customer’s question and your storefront.

It also converts differently than you’d expect. Reporting cited across several 2026 marketing guides puts the conversion rate of visitors who arrive via an AI search platform at something like 4 to 5 times higher than visitors arriving via traditional search. Fewer people are asking, but the ones who do are already halfway convinced — which is exactly why being the name the AI actually says matters more than it looks like it should.

How the model actually decides who to mention

This is the part that changes what you should do about it. AI answer engines pull from two different places, and they work on completely different timelines:

  1. What it already learned during training — this updates on a roughly 6-to-12-month cycle, so anything you fix here shows up slowly.
  2. What it looks up live on the web right now (retrieval-augmented generation, or RAG) — this refreshes far faster, typically within 24 to 72 hours of a page being published or updated.

Most small businesses that try GEO at all optimize for the wrong one — polishing an old “About Us” page and waiting months for nothing to change, when the faster win is making sure whatever’s live on your site right now is written in a way a model can actually lift and quote.

And once it’s looking, it’s not reading your whole page top to bottom the way a person might skim it. One 2026 analysis of AI citation patterns found that roughly 44% of what gets cited from a page comes from the first 30% of the text — the model is scanning for a clear, extractable answer near the top, not a well-earned reveal at the end of a long paragraph.

What actually moves the needle (no budget required)

Put the answer first, then the context. If a customer might ask “what’s the best [your category] for [specific situation],” your page should answer that specific question in the first sentence or two of the relevant section — not build up to it. A model deciding whether to cite you in half a second favors the page that hands it a clean, quotable sentence over the page that makes it work for one.

Name a real person behind the content. Visible author names, a real bio, and dates on your pages all feed into the trust signals these systems weigh — the same “who actually wrote this and when” signals search engines have used for years, just mattering more now that there’s no ranked list to hide an anonymous page inside.

Publish specific numbers, not vibes. “Fast shipping” is not citable. “Ships within 24 hours from [city]” is. A model can lift a concrete claim into an answer; it can’t lift a mood.

Get named on other people’s sites, not just your own. A model weighing whether to mention your business is also looking at whether anyone else talks about you — a review site, a local roundup, a supplier’s partner page. This is slower than editing your own site, but it’s the harder-to-fake signal, which is exactly why it counts for more.

Check what’s already being said about you before you optimize anything. Before spending an evening rewriting your homepage, actually ask ChatGPT, Gemini, and Perplexity the questions your customers would ask about your category, in your city, and see whether you show up at all — and if a competitor does, read exactly what the model says about them. That’s your actual starting point, not a guess.

What to skip

Don’t buy into “AI SEO audit” tools promising a ranking score — there’s no ranking here to score, only whether you’re mentioned or not in a given answer, which varies by exact question and by which model answered it. And don’t neglect your actual website to chase this: a RAG-based system is ultimately still reading real pages on the real web, so a slow, broken, or thin site hurts you here exactly as much as it always did — GEO doesn’t replace having a site worth citing, it just changes what “worth citing” looks like.

VERDICT

Don't hire an agency for this yet and don't wait for a bigger budget — spend one evening asking the three major chatbots the exact questions your customers ask, in your own city and category, and see who gets named. Then rewrite your two or three most important pages so the answer to a likely question sits in the first sentence, with a real name attached and a real number instead of an adjective. That single pass will do more than any amount of traditional SEO tinkering to make sure the next customer who asks an AI instead of Googling actually hears your name.

Sources