How AI search works

What has actually changed

For twenty years, finding a practitioner meant typing a few words into Google and choosing from a list.

That list is still there. What has changed is that a growing share of people now describe their situation in full sentences, to a machine, and get a single considered answer back instead of ten links to sift through.

Half of consumers now intentionally use AI-powered search, and Google places its own AI summary above the results it used to lead with. For a practitioner, this is not a new marketing channel to bolt on. It is a change in how the first conversation about you happens, often before you know it has happened at all.

The question is no longer only whether you rank. It is whether a machine can understand your expertise well enough to bring you up at the right moment.

From ten links to one answer

The difference matters more than it first appears.

A results page is a shortlist. It hands the reader ten options and lets them judge. An AI answer is a recommendation. It has already done the judging, named two or three sources, and moved on.

• A list rewards being present. An answer rewards being understood.
• A list gives everyone on page one a chance. An answer usually cites a handful.
• A list is the same for everyone. An answer is shaped by how the question was asked.

This is why a website that reads beautifully to a human can still be invisible to an answer engine. If the machine cannot tell what you treat, who you treat and why you are credible, it has nothing to work with, however good you are.

Being excellent at your work and being legible about it are two different jobs. AI search has made the second one matter.

How an AI answer is built

It helps to know what is happening between the question and the answer.

The exact mechanics differ between ChatGPT, Perplexity, Claude, Gemini and Google’s AI overviews, and none of them publish their workings. Broadly, though, four things happen:

1. The question is interpreted. A vague, human description is turned into what the system thinks is really being asked.
2. Sources are gathered. The system pulls from what it learned in training and, increasingly, from a live search of the open web.
3. Sources are weighed. It looks for material that is relevant, clear, consistent with other sources and credible enough to repeat.
4. An answer is composed. It writes one reply in its own words, usually naming only the sources it leaned on most.

Two things follow from this. First, most AI answers are drawn from the ordinary web, so the work of being findable has not been thrown away. Second, the step that decides your fate is the third one, and that is the step you can actually influence.

You cannot write the answer. You can make sure that when it is written, your expertise is in the material it is written from.

What makes a source get used

Across every platform, the same handful of qualities keep coming up.

• It can be read. The page loads, the content is in the HTML, nothing important is locked inside an image or a script.
• It is specific. “Holistic wellbeing” tells a machine nothing. “Somatic therapy for burnout in healthcare workers” tells it everything.
• It answers a real question. Content shaped around what people actually ask is far easier to lift into an answer than a page of atmosphere.
• It is consistent. Your name, qualifications, location and services should say the same thing everywhere they appear online.
• It is credible. Named authorship, real credentials, and other reputable sites referring to you all raise the odds of being trusted.

None of this is a trick. It is the same clarity that helps a nervous person on your website decide you are the right person to call, written down where a machine can also read it.

SEO, AEO and GEO

Three names, one continuous piece of work. They stack rather than compete.

You cannot skip to the top. A site that machines cannot crawl will not be understood, however clearly it is written; a site that is crawlable but vague gives an answer engine nothing to say about you. Each guide below goes into one layer properly.

Where to start

In practice, almost everyone starts in the same place, and it is not the exciting one.

• Make sure the site works: it loads quickly, reads properly on a phone, and nothing important is hidden from a crawler.
• Say plainly, in words, who you help and what you help with. Specifics beat atmosphere.
• Answer the questions people actually ask before they book, in their language rather than your profession’s.
• Make your credentials visible and consistent wherever your name appears.
• Then, and only then, look at the AI-specific layers.

This is deliberately unglamorous. It is also the part that compounds, and the part that keeps working whichever platform people are asking next year.

What nobody can promise

This field attracts a lot of confident claims. Very few of them survive contact with how these systems actually work.

No one can guarantee that ChatGPT will recommend you. AI platforms do not sell placement in their answers, do not disclose how sources are selected, and change their behaviour without notice. Two people asking the same question can get different replies.

What can be influenced is everything before the answer: whether your expertise exists in a form a machine can find, understand, trust and repeat. That is real work with real effects, and it is honest about its limits.

We can strengthen the conditions for you to be found. We cannot, and will not, promise the outcome.

Frequently asked questions