You Are Absent From the AI Answers. Now What?

You ran the test and the answer is zero. Not named, not cited, on every question that matters. The instinct is to treat that as a verdict on the business. It is not. A measured zero is the most actionable number this discipline produces, because it tells you precisely which questions you lose and whose evidence is winning them instead.

The short version

We have published our own zero three times: 21 July, 7 September and 12 September 2026, across 8 Dubai buyer questions, 0% named and 0% cited on every reading. Across the wider 62-query panel (16-18 September 2026) our Share of Recommendation reads 0.0, with 59 of 62 questions returning us absent. That zero is not a reason to stop measuring; it is the baseline everything else is judged against. An absence is diagnosable into four causes, each with a different fix, and the cheapest two can be done this month.

Read the absence before acting on it

Absence is not a penalty

Nothing in our data suggests engines punish sites. You are absent because the retrieval and generation steps selected other evidence, which means the response is to supply better evidence rather than to appeal.

Absence is specific

You are not "invisible to AI". You are absent from these questions, on this engine, on this date. That specificity is what makes it a work order.

Absence is common

Across our 62-query panel, 59 of 62 questions return us absent, 2 return us cited and 1 names us. A zero in a young category is the normal starting state, not an outlier. The panel itself shows why: across those questions the engine pulled 301 citations from 178 different domains, 131 of them cited exactly once, and the largest single source held 6%. The four platform surfaces hold 18% of the citations; our own domain holds 1.7%. There is no incumbent occupying the seat you want. The seat is shared by a crowd, and you are simply not in it yet.

The four causes and their sequence

Cause	Symptom	Fix	Cost shape
Entity unresolved	Engine describes you wrongly	Consistent identity, structured data	Low, fast
Not retrievable	Correct description, never in sources	Answer blocks, quantified claims	Low, fast
Not distinct	Described well, rival still named	Positioning, proof, an owned claim	Slow, highest value
Answer is off-domain	Sources are platforms	Participation and placement	Ongoing, rarely staffed

The sequence matters. Entity and retrievability are cheap and fast, so they go first even though they are not the most valuable. Distinctiveness is the expensive one and takes quarters. Off-domain is the one most teams never fund, and it is where almost half the answers get at least one of their sources: 27 of the 56 answers that cited anything cited YouTube, LinkedIn, Reddit or Quora.

What to do in the first 30 days

1. Fix resolution. Ask the engine who you are. If the description is wrong, correct your owned properties until it is right. Nothing downstream works before this.
2. Rebuild your three highest-intent pages at passage level. Answer block in the first 130 words, quantified claims, sources linked inline. The Princeton GEO research, Aggarwal et al., measured changes of this kind moving visibility by up to 40% under controlled conditions.
3. Classify every target question as on-domain or platform-held. Stop funding page work for the second group.
4. Pick one platform-held question and participate properly. Answer the actual thread the engine quotes, without promotion.
5. Book the second reading for day 60. Without it you cannot distinguish your work from the churn, and the churn is substantial: we measured a 100% turnover among agency leaders inside 52 days.

How to know whether it worked

Only a re-probe tells you, run to the same questions, engine and locale. Two numbers matter more than position: how many questions return you at all, and whether named and cited moved independently. A rise in citations with no movement in naming means your retrieval work landed and your distinctiveness work has not started.

What we are doing about our own zero

Our diagnosis is off-domain first, retrievability second. More than half the answers we want are assembled from platforms, so writing more of our own pages cannot reach them, and we have said so publicly rather than quietly sell the on-domain work. As Firoz Azees puts it: "a measured zero is a work order; a flattering estimate is a story you tell yourself." The next reading is scheduled, and we will publish it whichever direction it moves.

Questions buyers ask

How long before a zero becomes something else? Longer than any vendor will promise. Ours has held across 52 days and three readings, which is the honest benchmark we can offer.

Is a zero worse than a low score? It is cleaner. A zero tells you the consideration set does not include you. A low score mixes that with volatility.

Should I tell my board the number? Yes, with the sequence attached. A number without a work order invites panic; a number with one invites funding.

Does buying software fix it? Software reports the zero faster and in more places. It does not write the page or join the conversation, and the category's limits are worth understanding before the purchase.

What if my competitor is also at zero? Then the category is unclaimed, which is the best news in this article. Young categories are where coverage compounds fastest.

Which fix gives the fastest visible change? Retrievability, usually. It is engineering work with a short loop, and it is measurable at the next reading.

Where do I start if I have not measured yet? Run the test yourself. It takes an hour and produces the only thing that makes this article usable: your own dated baseline.

At Ivanooo, Firoz Azees runs Distinctiveness Engineering for the AI-answer era: measuring who the engines name, cite and recommend, then engineering the gap between listed and chosen. Start with a free AI visibility check.