Ask an assistant who does what you do. You are probably not in the answer.

Open ChatGPT and ask it to recommend a company that does exactly what you do, in your city. Then do it in Gemini, and in Perplexity. It takes four minutes.

Most people who run this test find one of two things. Either they are absent entirely, or they appear with details that are two years out of date. Both are worth knowing, and neither shows up in any dashboard you currently look at.

The shape of the answer changed

A search results page gives you ten links and lets you choose. An assistant gives you two or three names and a reason. There is no page two. There is no scrolling past the fold to find someone interesting.

That compresses the field brutally. Ranking eleventh on Google still gets you traffic from people who scroll. Being the eleventh most citable company for a question an assistant is answering gets you nothing at all, because the answer stopped at three.

What assistants actually reach for

They are not running your keyword strategy. Broadly they lean on three things:

• What they were trained on. A snapshot, often stale, weighted toward sources that get quoted a lot.
• What they can retrieve right now. Live search, which is why a crawlable, unambiguous answer on your own site matters more than it used to.
• What they can verify. Structured data, consistent details across the places that describe you, and corroboration from sources other than yourself.

The third one is where most companies quietly fail. Not because their content is bad, but because nothing about them is machine-legible. A business whose hours, service area, and specialisms exist only as sentences inside a hero image is not withholding information out of strategy. It just never occurred to anyone that a machine would need to read it.

The things that actually move this

Unglamorous, and mostly invisible to a human visitor:

• Structured data that describes what you are. Not the generic organisation stub a plugin emitted. The specific type, services, area served, and the details that let something answer a question about you without inference.
• A direct answer to every real question. If someone asks whether you serve a particular county, or work in a particular sector, or handle a particular certification, there should be a sentence on your site that says so plainly. Assistants quote sentences. They do not quote implications.
• Consistency across sources. Where your details differ between your site, your profiles, and a directory that scraped you in 2023, a retrieval system has to pick. It often picks wrong.
• Corroboration you did not write. Anything that repeats your claims in a source that is not you carries disproportionate weight, because it is the only thing a model can use to distinguish a real company from a confident website.
• A crawler that is not fighting your site. Content that only exists after JavaScript executes is content some retrievers will never see.

Being wrong is worse than being absent

Absence is a gap. An incorrect answer, delivered confidently, is a competitor.

We have seen assistants report service areas a company abandoned, pricing that has not been current since before a rebrand, and in one case a phone number belonging to a business that had closed. The person asking has no way to know any of it is stale. They just call the number.

Which is why monitoring belongs in this work. Not a rank tracker. Actually asking the questions your buyers ask, across the assistants they use, on a schedule, and recording what comes back. The answer changes without warning and nothing tells you.

Why this window is open

Search optimisation is a mature discipline. Whatever edge exists there has been ground down by people who have done it for fifteen years.

This has not happened yet. Most companies have not run the four-minute test. Most agencies are selling the same search retainer they sold in 2019 with the word AI added to the deck. The gap between knowing this matters and anyone systematically working on it is, for now, unusually wide.

That will close. It always does. But the test is free, and you should know what the machines are saying about you before you decide whether to care.