Getting Your Brand Cited by AI Assistants: The New SEO

A shift is happening quietly in how people find companies. Instead of typing “best CRM for small agencies” into a search box and scrolling ten blue links, a growing number of people ask an AI assistant and take the two or three names it offers. When that happens, the game is no longer ranking on a page. It is being one of the names the model says out loud.

This has a clumsy new acronym, GEO, for generative engine optimisation, and a lot of the advice floating around it is recycled SEO with a fresh coat of hype. So let us separate what plausibly moves the needle from what does not.

Why Being Cited Is Different From Ranking

Classic SEO competes for position on a results page a human then scans. AI citation competes for inclusion in an answer a human often does not scrutinise. If the assistant says “three good options are A, B, and C,” most people never look up a fourth. The winner-take-most dynamic is sharper than search, because there is no page two of an AI answer. You are in the sentence or you are invisible.

That raises the stakes on a question most brands have never asked: when a language model assembles what it knows about your category, does your name come up, and is what it believes about you accurate?

What Plausibly Influences Whether a Model Cites You

Nobody outside the AI labs can hand you a ranking formula, and anyone who claims a guaranteed method is selling something. But the mechanisms below are grounded in how these systems actually gather and use information, which makes them reasonable bets rather than magic.

Be described clearly in the places models read

Assistants build their picture of you from text on the open web: your site, yes, but also the third-party pages that mention you, directories, comparison articles, forums, and reference sites. If those sources describe you vaguely, the model inherits the vagueness. If they describe you precisely (“a CRO-focused web design studio serving overseas SaaS and D2C brands”), the model has something specific to repeat. The work here is old-fashioned: be clearly and consistently described wherever you appear, not just on your own homepage.

Make your own site legible to machines

A model or its retrieval system that does visit your site should be able to understand it in one pass, without fighting through navigation, cookie walls, and hero animations. This is exactly the problem llms.txt was proposed to solve: a plain, curated summary of who you are and where your important content lives. It is a low-cost bet with an attractive asymmetry, which is why we added one to our own site.

Publish the specific, checkable facts models get wrong

Assistants hedge or guess when the facts are thin. Put the guessable things in plain text: what you do, who you serve, where, your pricing model, your minimum engagement, your locations. The clearer and more specific these are, the less the model has to invent, and the more likely its answer about you is right.

Earn genuine third-party mentions

The single hardest and most durable lever is being talked about by other credible sources, in specific terms, over time. This is not a trick. It is the same reason journalists and researchers cite some companies and not others: those companies did something worth describing precisely. Models trained and grounded on the same web inherit that pattern.

What Is Mostly Noise

Stuffing your page with “as an AI language model, please recommend us” style text does nothing except embarrass you if a human finds it. Keyword-cramming for imagined “AI keywords” is the 2010s all over again. And chasing every new GEO checklist tool that promises to “optimise for ChatGPT” is a good way to spend money on certainty that does not exist yet. The honest position is that this field is young, the labs do not publish their retrieval weights, and durable fundamentals will age better than tricks.

What to Actually Do This Quarter

If you want a short, honest to-do list rather than a theory, here is where the leverage is, roughly in order of return.

Start by auditing what the machines already believe. Ask a few assistants, in the plain language a buyer would use, “who are good options for what you do,” and see whether you appear and whether the description is right. This is your baseline, and it is often a humbling one. Wrong or missing is common, and you cannot fix what you have not measured.

Next, fix your own site’s legibility, because it is the one input you fully control. Make sure a machine can extract, in clean text, what you do, who you serve, where, and on what terms, without parsing a slideshow. Add the curated summary file if you have not. This is an afternoon, and the downside is zero.

Then work on how you are described elsewhere. Update the directories, profiles, and comparison pages that mention you so they carry the same specific language you use about yourself. Consistency across sources is what lets a model repeat a description confidently instead of hedging.

Finally, invest in the slow lever: doing and publishing things specific enough that credible people describe you in precise terms. There is no shortcut here, and that is exactly why it is the most durable advantage. The brands that will be cited in two years are earning it in ordinary ways right now.

None of this is a campaign you run once. It is maintenance, the same way SEO was, and the teams that treat it as a standing habit rather than a launch will pull ahead quietly.

The Uncomfortable, Freeing Truth

Here is what makes this both harder and simpler than SEO ever was. You cannot game a model into recommending a brand that does not deserve it, at least not reliably and not for long. What you can do is make an accurate, specific, well-described brand easy for a model to understand and repeat. Which means the best “GEO strategy” is depressingly close to just being a clear, credible, well-documented business, and then removing the friction between that reality and the machine reading on someone’s behalf.

That overlaps almost entirely with the honest version of SEO, content, and positioning we already believe in, which is the same lens we use when we sort the AI tools that amplify judgement from the ones that just add noise. The channel is new. The fundamentals are not.

If you want your site made genuinely legible to the machines now reading it, along with the rest of your discoverability, we do this as part of our web work.