gr.AI
Question explained · gr.AI

What to ask before you hire a GEO expert

Seven questions, and all of them test the same thing: whether the GEO expert, agency or company in front of you measures or promises. They are written to work against any supplier, ourselves included, and further down we show where they catch us.

  1. 01

    Will you show me the baseline measurement?

    Without an initial state, any result at the end is a claim. The baseline is the one thing that cannot be manufactured afterwards.

    Good answer
    You get the numbers before you sign, or in the first few days, and you are told exactly how they were obtained.
    Red flag
    You will see the growth in traffic. Traffic rises and falls for ten reasons, none of which proves you appear in AI answers.
  2. 02

    Which questions did you ask, and on which platforms?

    Engine answers vary from run to run. A screenshot is not a measurement, it is a coincidence photographed.

    Good answer
    A list of questions, the number of repetitions, the platforms covered, and what exactly was counted: the appearance, the position, the citation with a link.
    Red flag
    A single ChatGPT screenshot, with no way of knowing what was asked, when, or how many times.
  3. 03

    Does your own site pass your own test?

    It is the cheapest check on the list and the hardest to fake: you can do it yourself, in five minutes, before the first conversation.

    Good answer
    They tell you straight where they themselves fall short, without being asked twice.
    Red flag
    They sell you AI crawler access while on their own site robots.txt allows the crawlers and the server blocks them. Checkable from the outside in two minutes.
  4. 04

    What do I end up owning?

    The difference between a report and a change on the site is who does the implementation, and with what access.

    Good answer
    You know from the start what you get as a document, what you get as a file ready to drop onto the site, what is done from the platform, and what needs access to the code.
    Red flag
    We will take care of everything, with no list of deliverables and no discussion of who has access to what.
  5. 05

    What can you not guarantee?

    The question that says the most about who is in front of you. An honest supplier has an answer ready.

    Good answer
    I cannot guarantee a position in a generated answer, because answers differ from run to run and the models change without notice. I can guarantee what I implement and I can measure what changes.
    Red flag
    A percentage. Any guaranteed percentage, offered without having measured anything on your side, is a number lifted from someone else's marketing.
  6. 06

    Will you give me named clients I can check myself?

    A logo on a website is not a reference. A company you can look up and ask is.

    Good answer
    Names you can verify, plus what exactly was done there.
    Red flag
    Anonymous case studies, with big percentages and a client in sector X.
  7. 07

    Who measures in three months, and with what?

    GEO is not an installation that finishes. Models change, competitors move, and a one-off measurement ages fast.

    Good answer
    The same methodology, repeated, with the figures placed side by side.
    Red flag
    Nobody. Or: a new measurement, with different questions, that cannot be compared with the first.

We applied the third question to the whole niche

The question about their own site can be checked from the outside, so we checked it on 27 Romanian agencies selling AI search optimisation services, in 2026-08-25. We put ourselves on the list too, under the same criteria. Another 2 could not be measured at all: one did not respond, one returned a server error, one filtered our connection. We do not give them a low score, but the fact that a visibility supplier's own site cannot be read is, in itself, an answer to the question above.

6/27block an AI crawler even though robots.txt allows it
25/27do not mark up their own reviews
10/27have no llms.txt

The first figure is the one that matters most: 6 out of 27 have a robots.txt saying they let AI crawlers in, while in reality the server blocks or degrades them. Exactly the mistake they sell as solved. 10 out of 27 have no llms.txt on their own site. 25 out of 27 ask clients to mark up their reviews with structured data without doing it for themselves, us among them.

Where we stand
90 out of 100, position 1 of 27, tied with one other. A shared position, not a title, and it changes at every rescan anyway. We are missing review markup. The figures, with every agency named, are published in full.

How we failed our own test
On the first run of this method, in 2026-08-24, our own site scored 75 out of 100. We fixed our own site with exactly the method we sell and the next day it scored 90. We publish both figures, because the second means nothing without the first, and because that is precisely the proof you should demand from anyone selling you AI visibility. Under version 1 of the method we had 93; that is not a drop, it is a different measurement. Version 2 also checks what the crawler actually receives, not only what the site declares.

What proves nothing

  • That they are the first in the country.

    Several claim the title at once and nobody checks it. Ask instead what they measured and when.

  • The number of blog posts.

    Content volume says nothing about whether an engine can read it. A site blocked at the crawler has exactly zero articles read, however many it published.

  • Certifications.

    There is no recognised certification for AI search optimisation yet. What exists are measurements, and those can be shown.

  • A screenshot of a good ChatGPT answer.

    Run the same question five times and you will get five slightly different answers. A screenshot is one run, not a result.

Share this article

Further reading