Measuring whether AI models mention your client’s brand

A growing share of buyers now ask a language model for recommendations before they ever open a search results page. For considered purchases, industrial equipment, professional services, software, that conversation often happens before anyone visits a website at all.

Which raises a question most agencies cannot yet answer for their clients: when someone asks an AI model about this category, does our client get mentioned?

It is answerable. Here is how we measure it and what we have found actually moves it.

The short version:

• One prompt in one model is an impression, not a baseline; run a prompt set across models
• Structure and third-party corroboration move mentions more than marketing copy
• Specification tables with explicit units outperform descriptive prose
• Nobody can guarantee AI placement, and anyone promising it is selling vapor

How to measure AI search visibility: build a defensible baseline#

The instinct is to ask ChatGPT a couple of questions, see the brand appear or not, and form an impression. That is not a baseline. These models are non-deterministic, so the same prompt produces different answers on different runs, and one lucky response tells you nothing.

What produces a usable number: build a set of prompts reflecting how buyers actually ask, run each one multiple times across multiple models, and record whether the brand appears, in what position, and in what context.

The prompt set matters more than the volume. Include the obvious category question, the comparison question, the specification-led question, the problem-led question where the buyer describes a symptom rather than a product, and the direct brand question to check what the model says when asked outright. That last one occasionally surfaces something alarming, like a model confidently describing a product line the client discontinued years ago.

Run each prompt enough times to smooth out variance, across several models rather than one. The output is a mention rate: a percentage you can track, report and improve. It also gives you the competitor picture for free, because you record who else appears.

What makes AI models mention a brand#

The consistent finding across the work we have done is that AI visibility is mostly downstream of structure and third-party corroboration, not of marketing copy quality.

Structure the model can parse. The pages that get quoted back tend to have clean heading hierarchies, explicit specification tables, and answers stated plainly near the top of the page rather than after three paragraphs of positioning. A page that opens by naming what the product is, who it is for, and what it does, in ordinary language, does better than one that opens with a brand promise.

Being described elsewhere. Models draw heavily on how a brand is characterised across the wider web. Industry directories, comparison articles, documentation, forums, and technical write-ups all feed this. A client with a beautiful site and no third-party presence is close to invisible, because nothing corroborates what the site says about itself.

Consistency of category language. If a client calls its product one thing and the market calls it another, models struggle to place it. This is a real and common finding. The fix is unglamorous: use the market’s word alongside the brand’s word, especially in headings and specification tables.

Specifications stated as data. For equipment and technical products, a specification table with explicit units is worth more than any amount of descriptive prose. It is the format that survives being extracted and repeated.

AI SEO tactics that do not show a measurable effect#

Adding an FAQ block to every page, purely for AI, has not shown a clear effect in the work we have measured. Neither has stuffing more of the brand name onto a page.

Schema markup helps, but as part of general structural clarity rather than as a lever on its own. A page with excellent schema and a confused heading structure does not outperform a clearly written page without schema.

Being wary here matters. This field is young, the measurement is noisy, and there is a lot of confident advice being sold that nobody has tested. We would rather report a finding as uncertain than present it as settled.

Selling AI visibility as an agency service#

The measurement itself is a sellable deliverable, because it answers a question the client’s competitors also cannot answer. A baseline report showing mention rate against named competitors, with the prompts and methodology included so it can be re-run, is a genuinely useful artefact.

From there the work is mostly content and structure, which is work agencies already know how to sell: rewriting key pages to lead with plain answers, building specification tables, producing comparison content that addresses the questions buyers are actually asking, and pursuing the third-party presence that gives models something to corroborate against.

Re-run the baseline quarterly. The number moves, and being able to show it moving is the difference between an interesting one-off report and an ongoing engagement.

Why guaranteed AI rankings are a red flag#

Nobody controls how these models rank or select brands, the systems change without notice, and anyone offering guaranteed placement is selling something they cannot deliver.

What you can do is measure where a client stands, make their information as easy to parse and corroborate as possible, and track whether it improves. That is a defensible service. Guaranteed AI rankings are not, and an agency that promises them will be having a difficult conversation within two quarters.

Frequently asked questions#

How do you check if ChatGPT mentions your brand?#

Not by asking it twice. Build a set of prompts reflecting how buyers ask, run each multiple times across several models, and record whether the brand appears and in what position. The output is a mention rate you can track quarter over quarter.

Does schema markup improve AI search visibility?#

It helps as part of overall structural clarity, not as a standalone lever. A clearly written page with plain headings and specification tables outperforms a confused page with excellent schema.

Can an agency guarantee AI search rankings?#

No. Nobody controls how these models select brands and the systems change without notice. What can be delivered honestly is a measured baseline, structural improvements, third-party corroboration work, and a tracked mention rate over time.