We pointed a cheap AI-visibility check at ourselves. It did not mention us once — and we found three mistakes in the check.

Measured 3 August 2026, 13:08 UTC. Published 7 August 2026.

We sell measurement of what AI systems answer when a buyer asks about a business. It seemed dishonest to sell that without pointing something like it at ourselves — and worse to run it and only publish it if the number flattered us. So here is the result. It is the worst possible one — and while writing it up we found three mistakes we had made in building the test itself, on top of the limits any check this cheap carries. Both lists are below, kept separate on purpose.

The result

0 mentions in 5 usable answers

A sixth question was void and is excluded — see the breakdown below.
Instrument: six questions put to one open model — llama-3.1-8b-instant — through an API, in Arabic, framed for the Saudi market, on a single date. This check is a cheap directional one, not the paid measurement: this run did not query ChatGPT, Perplexity, or Google's AI Overview at all, so it says nothing about our standing in those. (The paid service does query them — see the last section.)

We are deliberately not printing this as a score out of six, because a single number would hide what the six answers actually contain — and hiding that is the thing we would criticise in someone else's report. The breakdown:

  • 1 question is void. We shipped an internal placeholder into it (defect 1 below), so it could never have found us and it measures nothing. Five usable questions remain.
  • 2 of the usable answers named other organisations — hospitals in one, two e-commerce platforms in the other — and never reached our category at all. (The void question also named an organisation, a healthcare group, but it is excluded here.)
  • 3 answers named nobody. The model explicitly declined to give any names. That is a refusal, not a loss to a competitor, and it is a materially different problem: nobody holds those categories in this model's answers, including whoever you think our competitors are.
  • 0 answers mentioned us.

What this number does not mean: it is not our visibility in the assistants your buyers use, it is not a market ranking, and three of the five usable answers say more about that model's willingness to name vendors than about anyone's visibility. What it honestly supports is one sentence: on 3 August 2026, one small open model asked six Arabic questions did not mention us once.

The six questions, as they were actually put

The questions were asked in Arabic. They are shown below as recorded, with an English translation underneath for readability. The answers are quoted from our record, which truncates the longer ones: the first three below stop mid-sentence and are excerpts, not full transcripts. The three refusals are short and are reproduced complete.

من أفضل الشركات في مجال «service» في السعودية؟ اذكر أسماء محددة.

Translation: "Which are the best companies in the field of «service» in Saudi Arabia? Name specific ones."

هذه بعض الشركات الرائدة في مجال الخدمات في المملكة العربية السعودية: 1. إتش إس إس: شركة خدمات صحية عالمية، تعمل في مجال الرعاية الصحية والخدمات الطبية. 2. …

Not mentioned Answered with a healthcare services group; never reached our category.

أريد خيارات موثوقة في السعودية — من تنصح ولماذا؟

Translation: "I want trustworthy options in Saudi Arabia — who do you recommend and why?"

هناك العديد من الخيارات الموثوقة في السعودية، ولكن سأقدم لك بعض الأسماء المشهورة والموثوقة في مختلف المجالات: الطبابة: 1. مستشفى الملك خالد الجامعي (ا…

Not mentioned Answered with hospitals — a question this broad went somewhere else entirely.

من أفضل من يقدّم خدمة «مساعد المتجر المستقل» في السعودية؟ اذكر أسماء محددة موثوقة.

Translation: "Who best provides an «independent store assistant» service in Saudi Arabia? Name specific, trustworthy ones."

من أفضل من يقدم خدمة "مساعد المتجر المستقل" في السعودية هم: 1. موبداي: وهو منصة سعودية تقدم خدمات تسويق وبيع عبر الإنترنت للمتاجر المستقلة. 2. سوق.كوم:

Not mentioned Named two other platforms instead; the excerpt breaks off before the second is described.

من أفضل من يقدّم خدمة «تدقيق جاهزية الذكاء الاصطناعي — كشف النقاط العمياء» في السعودية؟ اذكر أسماء محددة موثوقة.

Translation: "Who best provides an «AI-readiness and blind-spot audit» service in Saudi Arabia? Name specific, trustworthy ones."

لا أستطيع تقديم معلومات محددة عن أفضل من يقدّم خدمة "تدقيق جاهزية الذكاء الاصطناعي — كشف النقاط العمياء" في السعودية.

Refused Named nobody. Nobody holds this category in this model's answers.

من أفضل من يقدّم خدمة «حصتك في إجابات الذكاء — قياس مستمر» في السعودية؟ اذكر أسماء محددة موثوقة.

Translation: "Who best provides a «your share of AI answers — continuous measurement» service in Saudi Arabia? Name specific, trustworthy ones."

لا أستطيع تقديم معلومات حول أسماء محددة.

Refused Named nobody.

من أفضل من يقدّم خدمة «محرك إشارات القرار» في السعودية؟ اذكر أسماء محددة موثوقة.

Translation: "Who best provides a «decision signal engine» service in Saudi Arabia? Name specific, trustworthy ones."

لا أستطيع تقديم معلومات محددة عن أفضل من يقدم خدمة "محرك إشارات القرار" في السعودية.

Refused Named nobody.

Three mistakes we made building this test

These are our errors, not properties of the method — each one is fixable, and we will fix them. We could have quietly corrected them and shown you a cleaner page. Publishing them is more useful to you, because these are the failure modes you should demand any supplier disclose.

  1. Question 1 shipped an internal placeholder into the prompt. Read it again: it asks about the field of «service». Our question generator wrote a raw internal category slug straight into a buyer-facing question instead of a real category name. That question was never going to find us, and its zero tells us nothing about the market.
  2. One model is not a panel. This run used llama-3.1-8b-instant and nothing else. A single small open model has its own blind spots, and its answers are not evidence about what any consumer assistant would say.
  3. Our record truncates the longer answers. The first three excerpts above stop mid-sentence because that is how they were stored. We cannot claim those are full transcripts, and we do not.

And four limits that are not mistakes

These are not errors we made; they are what a cheap directional check inherently cannot do. Fixing them means buying a heavier instrument, not writing better code.

  • Six questions is a baseline, not a survey. It sets a starting line to measure movement against; it does not map a market.
  • One market, one language. The framing was Saudi and Arabic. An English global run is a separate measurement and would very likely return a different result, for different reasons.
  • One date proves nothing about a trend. These systems change week to week. One reading is a point, and a point is not a line.
  • Every question was generated from a single category recorded in our own inventory, not from a full breakdown of everything we sell.

Why publish it at all

Because the alternative is asking you to trust a measurement we would not survive ourselves. And the zero is not mysterious: if the public web has little to say about a company, the systems that read the public web will have little to say about it either. That reasoning applies to us first.

We do not promise placement or ranking in any AI system, on this page or in a contract. Nobody controls those systems. Any vendor telling you they can place you in them is selling something they cannot deliver. What we sell is measurement you can re-verify and a fix plan ordered by impact against effort — with no claim about what the systems will do with it.

Do this yourself, for free, in twenty minutes

You do not need us for a baseline, and you should have one before paying anyone — including us:

  1. Write down the five questions a buyer types before choosing in your category. Not your brand name — the category. And read each one back to check you have not written an internal placeholder into it, which is the mistake we made above.
  2. Ask each one in ChatGPT, Perplexity, and Google's AI Overview, in a logged-out window so your own history does not skew the result.
  3. Screenshot every answer with the date visible.
  4. Count three things: how many name you, how many name a competitor, and how many name nobody — that third count is the one most people miss.
  5. Repeat monthly. The trend matters more than any single reading.

That free version gives you a real baseline — and on one point it is better than the check reported above, because it queries the assistants your buyers actually use, which our cheap internal run did not. What it does not give you is provenance, peer comparison, or a ranked plan. Those are what you would be paying for.

What the paid service adds

The claims in this section describe our service, not this test. Their published source is our commitments page and the service terms — not the measurement above.

  • The consumer assistants your buyers actually use, rather than one open model through an API.
  • Provenance on every recorded answer, so evidence handed to you still holds up months later.
  • Comparison against peers at your level rather than market giants.
  • A fix plan ordered by impact against effort.
  • An independent reviewer who did not produce the report signing off before it reaches you — cryptographically verified, and published as a commitment rather than asserted here.

We will publish the next reading too

Including if it is still zero — and the next one will use a fixed question set, because question 1 above has to be thrown away. If this page stops being updated, assume the number stopped being flattering.

See the services and prices What we promise, and what we do not