When AI Cites You Wrong: Why Presence Is Not Enough

AI cites you, but the customers who come from there are always looking for the lowest price — far from your positioning. It is no accident: if your online mentions live inside articles about cheap deals and the mass market, the model learns that you are a brand for people who want to spend little. Every citation in the wrong context is worth as much as no citation at all. Changing the context in which you appear is simpler than it seems — and it can turn that channel into a source of truly on-target customers.

Organic traffic holds steady, PR mentions roll in, your brand even shows up in Perplexity’s answers. Yet the quote requests that come through AI are always from the wrong segment — people asking for discounts, not customers aligned with the premium positioning you have built. There is a hidden reason, and it lies in the context in which AI learned you.

Let me explain what I have been seeing for months as I compare brands in the same sector. The bare mention, detached from the context in which it lives, is not a signal of strength. Two artisan shoemakers from Le Marche can both be cited by Perplexity when you ask for “best Italian artisan shoemakers”. One comes out as a benchmark of excellence. The other comes out as an example of the “mid-market”. Same region, same supply chain, same Goodyear construction. But the context in which AI learned them is different — and that context weighs far more than the fact of being cited.

In the previous articles in this series on AI citations I have been dismantling the idea that “just being mentioned is enough”. Here I show you why the context of the citation is the real signal, and what to do when you negotiate a PR mention for your brand.

What the research says about citation quality

Huang et al. (2024) distinguish two different planes when an AI model produces an answer accompanied by sources: on one side the correctness of the information returned, on the other the quality of the link between that information and the cited source. These are two axes evaluated separately, not a single thing.

From this follows a deduction that the paper does not make explicitly in terms of brands, but which is direct for anyone working on visibility: if the quality of the citation-information link is evaluated as a standalone dimension, then the context around the mention — that is, the host piece inside which your brand appears — becomes part of the signal the model learns. It is not enough for your name to be present. What counts is how that presence fits into the semantic perimeter of the document that cites you.

The operational consequence for your brand is bigger than it seems. If the mention of your shoemaking business comes inside an article whose central theme is “cheap clones of Blake shoes”, AI absorbs one thing only: your brand belongs to that context. It does not matter that the text says “of excellent quality”. The perimeter of the piece drags you along.

Why context comes upstream of everything else

If you have read the previous articles in the series, you know that AI does not reason by keyword but by relationships between entities. I talked about this in author entity recognition and in the inverted pyramid. A citation is a relationship between your entity and the entity “topic of the host piece”. That relationship has a sign, a weight and a direction.

The implicit weight of the mention — the way AI interprets it — I described in implicit reference weight. Here I add the missing piece: the weight depends on the narrative role that the host piece assigns you. Leader, follower, budget alternative, edge case, example of error. These are five different roles and AI tells them apart.

The A/B comparison I ran with two shoemakers from Macerata

Let me take you inside a test I wrapped up at the end of March. Two artisan shoemakers from the Macerata district, I will call them Brand A and Brand B. Similar price range (men’s shoes 300-450 euros), same Goodyear construction, same channel (high-end multi-brand retail + small direct e-commerce).

I collected all the web mentions from the last 18 months: 14 for Brand A, 11 for Brand B. Close numbers. Then I classified each mention by narrative role:

• Brand A: 9 mentions out of 14 in pieces where it was cited as a “benchmark of artisan Made in Italy”, anchored to workshop stories, third generation, ties with historic cobblers.
• Brand B: 7 mentions out of 11 in comparative pieces with headlines like “accessible alternatives to the big labels” or “where to find Italian shoes without spending 800 euros”.

Then I ran 12 queries on Perplexity and ChatGPT with prompts like “top artisan shoemakers in Le Marche”, “who makes the best Goodyear shoes in Italy”, “where are hand-stitched shoes made in Le Marche”. Across the 12 queries, Brand A appeared in 8 answers with descriptors like “benchmark”, “artisan excellence”, “historic workshop”. Brand B appeared in 5 answers, always with descriptors like “more accessible option”, “alternative”, “good value for money”.

An indicative test, not a controlled study. Small sample, clear pattern. Brand B is not “invisible” — it is visible in the wrong role relative to the positioning the owner had described to me (he wanted to be perceived as top of the range). AI is teaching it, every time it gets cited, that it belongs to the “mid-market”.

I state the limits: I do not know how much each individual host article weighs, AI answers vary over time, 12 queries are not a complete audit. Serious analysis requires professional tools and systematic work on the sources.

The test you can run yourself in 20 minutes

Take your brand and do this round, without paid tools:

1. Open Perplexity and run 5-8 queries about your sector with different intents: “best X in [region]”, “top Italian X”, “cheap alternatives to [top brand]”, “historic X companies”. Use your customer’s language, not your own.
2. When your brand appears, read the full sentence in which it is cited, not just the name. Note the descriptor (leader, alternative, historic, emerging, budget).
3. Click on the sources Perplexity cites below the answer. Read the headline and the first paragraph of the host piece. What is the overall narrative role of that piece?
4. Do the same exercise with 3 competitors that AI cites alongside you.

Binary decision threshold: if in 5 mentions out of 8 the role AI assigns you matches the positioning you want, you are fine. If it matches in fewer than 3 out of 8, you have a problem of citation context, not quantity.

The mistakes I see most often when PR mentions are negotiated

I often work with marketing managers who have a PR budget and use it badly relative to AI. The patterns that keep recurring:

• Chasing the outlet, ignoring the piece. “We got cited in [major business outlet].” Great, but in which piece? If the piece was “the Italian districts that are struggling”, the mention taught AI the wrong context.
• Accepting weak descriptors. “Shoemaker from Macerata” is weak, “third-generation Goodyear workshop in Montegranaro” is strong. Same company, two different semantic relationships. AI learns the difference.
• Not reading the host piece’s headline before approving the mention. The headline is the strongest signal of the narrative role. If the headline is a downward comparison, the mention inside it does not save you.
• Treating all mentions as equivalent in the PR report. In your press office’s internal report, the “context of the mention” column almost never exists. Add it.

What to do concretely starting tomorrow

Coming back to the reasoning on attributable generation in Huang et al. (2024), the logical point is this: if AI models evaluate citation quality as a separate axis from mere presence, then the coherence between your brand and the context of the sources in which it appears becomes an active lever. It is no longer a bonus, it is part of the signal.

In practice, these are the actions:

• For every mention you negotiate, ask to review the descriptor before publication. Move from generic to specific-and-positioning.
• Build a one-page media kit with 3 descriptors you want AI to learn. Always send it to the journalist or the blogger.
• Map the 10-15 mentions that already exist and flag those with an incoherent role. Where possible, ask for a correction or a supplementary note.
• When you do guest posts, the host piece’s headline counts more than the mention inside it. Negotiate that too, not just the link.

What this has to do with your visibility in AI answers

The context of citations is the lever that turns the quantity of mentions into quality of the positioning AI learns. If you stop at the count, you are doing 2018 digital PR. If you read the role, you are working on visibility in 2026 AI answers.

In the next articles in this series I get into the detail of citation velocity (the pace of mentions over time), mention sentiment (how the tone of the citation is read by the models) and reference clusters (when several sources cite you together with other specific entities). Three faces of the same coin: it is not enough to be there, what counts is how you are there.

If you want to tie the thread back to the other topics, two reads that hold up the argument are backlinks as a citation proxy and E-E-A-T for AI. The first shows you why the traditional link is no longer the right KPI. The second tells you what the right KPI is.