—
For about two years, the pitch for AI visibility work went like this: AI answers can’t be bought, so the only way in is to earn it.
Then OpenAI started testing ads inside ChatGPT in February 2026, first for US users on the Free and Go tiers. Google had already been placing ads inside AI Mode answers. Within a few months, a new category of tools appeared just to monitor AI ad placements.
The obvious reaction in a lot of marketing meetings: “Great, so we can just pay for it now.”
The data says otherwise, and it’s the strongest argument yet for taking LLM visibility tracking seriously.
Paying for the slot doesn’t buy the answer
SE Ranking analyzed around 50,000 commercial keywords this summer and found ads inside Google AI Mode answers on nearly 30% of them, with expensive keywords far more likely to carry an ad.
The more revealing number came next. When an advertiser’s ad appeared in an AI Mode answer, that same advertiser was cited as a source in the answer only 11.5% of the time.
OpenAI has drawn the same line from the other side, stating that ChatGPT ads are labeled and kept separate from the response itself. Running an ad doesn’t change whether the model mentions you organically.
So brands now have two distinct ways to appear in an AI conversation:
- Paid placement: a labeled ad unit next to the answer. Bought, targeted, reported in an ad platform.
- Organic presence: being named, recommended or cited inside the answer. Earned through content, reviews, mentions and authority.
They run on different systems, move for different reasons, and need different measurement. The problem is that most teams can’t measure the second one at all. Semrush’s AI Visibility Index, built on 126 million prompts, found that 45% of marketing leaders can’t accurately measure their brand’s visibility in AI answers.
Add paid placements to that blind spot and the picture gets murkier. A sales rep sees your brand in a ChatGPT session and reports “we’re showing up in AI.” Was that the ad or the answer? Without tracking, nobody knows.
What LLM visibility tracking actually measures
Classic rank tracking watches one list of ten blue links. LLM visibility tracking watches something messier: generated answers that change by prompt, by model, and sometimes by the day.
A useful setup records five things for every tracked prompt:
- Did your brand appear in the answer at all?
- How were you described? Recommended, listed neutrally, or mentioned with a caveat? Does it match your positioning?
- Citation source. Which page did the model cite when it mentioned you: your site, a review platform, a Reddit thread, a competitor’s comparison post?
- Who appeared alongside you, and who appeared instead of you?
- How all of the above moves week over week, because a single snapshot of a probabilistic system tells you very little.
Citation source is the one I’d underline. It tells you what is actually earning the mention, which is the closest thing AI search has to a ranking factor you can act on.
How to build a prompt set that reflects real buyers
The quality of your tracking depends almost entirely on the prompts you choose. Three practical rules:
- Map prompts to buyer stages. Awareness (“what is AI visibility?”), consideration (“how do teams track brand mentions in ChatGPT?”), and decision (“best tool for tracking AI citations for an agency”). Decision-stage prompts are where ads cluster, so they’re where organic tracking matters most.
- Use the language buyers use. Pull phrasing from sales calls, support tickets and the questions prospects ask in demos, not from your keyword list.
- Track the same prompts across models. ChatGPT, Perplexity and Gemini cite very differently. Semrush’s index found ChatGPT citing around 15 sources per answer against roughly 3 for Gemini, so a brand can be visible in one and absent from another.
Start with 20 to 30 prompts you’d be embarrassed to lose, run them weekly, and expand once the trend lines settle. If ChatGPT is your priority, this guide on how to track ChatGPT brand mentions walks through a manual baseline before you automate it.
Keep paid and organic in separate columns
If your team is testing ChatGPT ads, set up reporting with three distinct lines from day one:
- Organic AI presence (mentions and citations inside generated answers)
- Paid AI placements (ad impressions and clicks from the ad platform)
- Traditional search (organic and paid, as before)
The reason is simple. If organic AI presence drops while ad spend rises, a blended number will look flat, and you’ll find out about the drop months late, when the ads stop running.
A practical sequence for most brands: baseline your organic visibility first, then test ads on the high-intent prompts where you have no organic presence yet. Paid fills gaps. It shouldn’t hide them.
What to do with the data
Tracking only matters if it changes what you publish. The patterns I see most often:
- You’re absent and a review site is cited. Your gap is third-party proof, not your own blog.
- You’re mentioned but not cited. The model knows your brand but isn’t sourcing your pages. Look at structure, freshness and whether your pages answer the prompt directly.
- A competitor wins one model but not others. Study what that model cites for them. It’s usually a specific page type or community thread.
- Your description is off. Old pricing, a discontinued feature, the wrong category. Fix the source pages the model leans on.
The open questions
AI ads are going to get better targeting and better reporting, quickly. That will make paid AI presence easy to measure and organic presence comparatively invisible, unless teams build the tracking now.
What I haven’t settled yet: will ad placements eventually influence which brands models treat as relevant, even if the systems are separate today? How should agencies report organic AI visibility to clients who only understand rankings and clicks? And as answers personalize further, does a fixed prompt set stay a reliable proxy for what real buyers see, or will tracking need to sample by persona the way paid targeting already does?
—
