A day at Semrush's AI search bootcamp, and the number that reframes the job

I spent today at Semrush's AI search and brand visibility bootcamp. The useful part was not a tactic. It was a number: the proportion of sources an AI answer pulls from your own website is small, and it was put on stage as somewhere between five and ten per cent.

If that is even roughly right, then most of the work that decides whether an AI assistant recommends you happens on pages you do not own. That is a different job from the one most of us have been selling, including us.

When someone asks an assistant for the best accountant in Bristol, or the best help desk for an e-commerce business, the model retrieves a set of sources and builds an answer from them. The session's point was about the composition of that set. Your own site is in there. It is a minority of it.

The figure given on stage was five to ten per cent from brand-owned sources. Later in the day someone put it closer to fifteen. I am reporting the range rather than picking the version I like, because the honest position is that it was a figure in a presentation rather than a study I can hand you.

The direction is what matters and the direction was not ambiguous all day. The majority of what gets cited about you is written by someone else.

The framing I will steal is that your site is the source of truth, not the source of visibility.

Somebody told a story about looking for a Vienna restaurant that could seat twelve for a family party. They asked an assistant, got a list, and then, being in this industry, went and worked out why those particular restaurants. Two things were true of each of them. The restaurant's own site had a page saying plainly that it is a good place for large groups, parties and celebrations. And the Google reviews said the same thing back, in customers' own words.

The site stated it. Other people corroborated it. That combination is what got recommended.

That is the clearest version of this I have heard, and it is worth testing against your own business tonight. Is the thing you want to be known for written down plainly on a page? And is anyone else saying it?

If the answer to the first is no, the second will never happen. If the answer to the first is yes and the second is still no, you have a different problem, and it is the one most of the day was about.

Semrush teaches it as four, and the ordering is the argument.

Clarity. What you are, what topics you own, said the same way by everyone in the business. The day opened with this deliberately, and the reasoning is that everything after it compounds only if it stays consistent. Most of the room's hardest problem was not technical, it was getting an organisation to agree what it is.

Discoverability. Your own website. The part closest to classic SEO and the part everyone in the room already controls. This is where structure, speed and plain factual pages live, and it is the subject of our guide to what every page needs.

Authority and trust. Everywhere else. Reviews, articles, communities, comparison pages, the places people argue about you. Authority is whether you are credible on a topic. Trust was defined as whether a model will recommend you without hedging, which I thought was a sharp way to put it. You have seen the hedging: "some sources suggest".

Then the measurement and the cross-functional argument about who owns any of it, which is where most businesses actually get stuck.

The order is deliberate. You cannot build corroboration for a position you have not decided on.

The line I most agreed with: authority is not a single placement.

It is one topic, referenced consistently, across a lot of different sources, over time. Not one editorial mention in a good publication. The example given was a brand whose own website is weak and whose customers have been enthusing about it on Reddit for years, to the point where an assistant is now its largest source of new business.

We stopped selling monthly backlinks because our own profile turned out to be mostly footer credits from sites we had built, which is a receipt rather than a link profile. This is the better version of the argument, and it is the same conclusion from a different direction: what counts is being talked about on the subject you want to own, repeatedly, in places you do not control.

The practical consequence came up more than once. The people who used to do link building outreach are now doing brand mention outreach. Same skill, different target, and nobody in the room had a settled name for the job yet.

The second consequence is less comfortable. A list of target sources goes stale the way a keyword list goes stale. Sources vary by platform and vary a lot by industry, so it needs rechecking rather than writing once.

Carefully, and not with one number.

Ask an assistant the same question twice and you will often get two different answers. Ask from a different account and you probably will. So a single screenshot is a screenshot, and a single visibility score implies a stability that is not there.

The approach I would defend, and the one we use, is duller. Fix a set of real buyer questions. Ask each one several times on a schedule. Count how often you appear and turn that into a frequency rather than a position. Watch the trend across a month. Report it as an observation.

We wrote this up before the bootcamp rather than after it, in how to measure AI visibility honestly, and nothing today changed my mind about it.

1. 01Write down the three topics you want to be known for. Not services. Topics. If two people in your business would answer differently, that is the job for this week.
2. 02Check each one is stated plainly on a page you own, in a sentence a machine could lift without interpreting you.
3. 03Then go and find out who else says it. Search for your topic and read what comes back about you. Reviews, forums, listicles, comparison pages. That set is your actual visibility, and most of it was written by someone who has never spoken to you.
4. 04Pick the five sources you could realistically influence and work out who to contact. Not a hundred. Five.
5. 05Fix your questions and ask them repeatedly, so that in three months you have a trend rather than a screenshot.

Whether the five to ten per cent holds across industries, or whether it is much higher for businesses with little written about them. A local trade with no Reddit presence and no listicles might well be mostly its own site by default, which would make the advice almost the opposite.

Nobody asked that, and I wish I had.

The other one: everything in this model rewards being talked about, and being talked about is easier if you are already large. The honest answer for a small business is probably reviews, because it is the one third-party source almost anyone can build, which is what we wrote about reviews earlier this year.