AI Choosing You Free scan

GEO and AEO services

Done-for-you GEO: the weakest company an AI assistant named in our runs had one referring domain. Authority is not the entry ticket.

We measured the off-page strength of 9 companies an assistant actually named. The range ran from 1 referring domain to 2,095. If you have been told this work starts with a link-building budget, that is not what the named set looks like.

1

referring domain on the weakest company an assistant named, against 2,095 on the strongest

What we measured

  • Among 9 companies an AI assistant named, off-page strength ranged from 1 referring domain to 2,095 - a spread of more than 2,000 times.
  • The weakest company in the named set had 1 referring domain, 1 backlink and 20 organic keywords, and was named anyway.
  • 5 separate explanations for who gets named - page structure, sentence matching, a page per trade, site shape and off-page authority - all failed to predict it in our tests.
  • Across 3 keyword families totalling about 1,160 trade-specific phrasings, 387 of 387 and 276 of 276 registered no measurable search volume - a page per trade is not what buyers search for.

What we ruled out, so you do not pay for it

5 page-side and domain-side explanations were tested separately. All 5 came back negative.

Structure: across 112 cited and 128 uncited pages answering the same queries, cited pages were shorter, carried fewer numbers, and had less structured data - 70% against 80%. Sentence matching: a cited page's best-matching sentence beat an uncited one in 10 of 26 comparisons, below a coin flip. A page per trade: 3 keyword families totalling about 1,160 trade-specific phrasings returned no measurable volume, 387 of 387 in one and 276 of 276 in another. Site shape: the 9 named companies ranged from 24 to 565 pages, and only 3 of the 8 we could judge had a page for the trade they were named for. Authority: 1 referring domain to 2,095.

None of these is worthless for other reasons. None of them is the mechanism, and none should be priced as if it were.

What a service should actually be doing

What the cited pages had in common was subject matter. On one pricing question, every cited source was a page containing actual figures - a pricing survey, a cost calculator, a rates directory. Not the best-designed pages: the pages that owned the number the answer needed.

So the work is to find the questions a business's customers actually put to an assistant, find which facts those answers require, and make that business the source of them. Then read repeatedly to see whether it worked, because a single reading cannot tell you.

That is slower than a markup checklist and it is the part that cannot be faked by a competitor who has not done it.

The honest boundary of the offer

We make a site citable. We do not promise placement in any assistant's answer, and nobody who has measured this surface honestly can.

Our own pre-registered baseline says the same thing about us: on the questions in this category, our own site is currently cited zero times. We wrote down the threshold for judging our own pages before we built them, and we will publish the reading either way.

The questions this page answers

  • generative engine optimization geo services720/mo
  • answer engine optimization services480/mo
  • answer engine optimization agency170/mo
  • leading generative engine optimization services in ai industry140/mo
  • best ai driven visibility optimization services90/mo

5 questions, 1,600 searches a month between them. All 5 answers we examined on them were produced by a web search, which means a page can reach every one.

Who an assistant cites on these questions today

  • youtube.com 4
  • firstpagesage.com 2
  • greenbananaseo.com 2
  • tryprofound.com 2
  • powerdigitalmarketing.com 1
  • 1digitalagency.com 1
  • cazarin.com 1
  • geekytech.co.uk 1
  • mainstreethost.com 1
  • perrill.com 1

35 distinct domains are cited across these questions; the 10 above are the most frequent, with the number of answers each appears in. Counted from the answers captured in our corpus, not from a market census.

What to ask before you buy

  • Ask what the vendor measured before proposing work, and ask to see the raw answers. A proposal with no reading behind it is a template.
  • Be sceptical of any offer priced around links, schema or a page per trade. All 3 failed to predict naming in our tests.
  • Ask what the vendor will publish if the work does not move the number. We publish ours.

Honest limits of what we measured

  • 9 companies, no control group, third-party estimates for the off-page figures and one point in time. Enough to show authority is not required; not enough to say what it is worth.
  • The negative results are negative results at modest samples. Absence of an effect at this scale is not proof that no effect exists.
  • We do not measure or diagnose classic Google rankings, and nothing here should be read as advice about them.

Questions we get asked

Do you work with agencies?

Yes. The measurement is identical whether it is run for one business or for a book of clients, and the report can carry your branding. Terms for multiple clients are a conversation rather than a price on a page.

What does a first engagement look like?

A reading before any work: the questions your customers would ask, the raw answers, who is cited today. Only then a proposal, because until the reading exists there is nothing to propose against.

Can you guarantee we get named?

No, and we would not trust anyone who does. Our own repeats of one identical question overlapped 15% on average - the set moves. What we can promise is a measurement you can check and a rate you can watch.

How long does it take to see a change?

We do not know, and we will not invent a figure. There is no measured time-to-citation in our data - the publication dates were empty on all 1,935 answers in the corpus we read. We measure at intervals after the work and report what we see.

Start with one free reading

One business, real questions, the raw answers back - before you decide on anything paid. If you are running this for a book of clients, say so and we will talk terms rather than quote a page price.

Free reading

Takes about a minute.

Free · no card · we keep what you type only to run the scan and email the result.

Related

Pricing is per category, because the work differs by how competitive the answers are. See every category we measure and what a reading costs →

Published 31 August 2026. Every figure on this page is recomputed from raw observation files by a single script, so any of them can be traced back to the readings behind it. We make your site citable. We do not promise placement in any assistant’s answer - we measure whether you are in it, and show you the raw answers either way.