I'm still not sold on prompt tracking as a measure of how people actually find and choose products.
A big part of that thinking comes down to what happens after the first search, and what actually triggers that search.
Someone might start with “family SUV”.
AI asks about their budget.
“$50k.”
Whether they'd consider a hybrid.
“Yes.”
Read those last two replies as standalone queries and much of the decision context is missing.
That's part of what I'd call secondary search. The follow-up exchanges with LLMs where someone adds their own context, circumstances and objections. It's my term for this behaviour, rather than an official Google label.
I've seen “yes” appear in Search Console against product pages. That doesn't show the conversation behind it, but Google confirms that AI Mode follow-ups count as new queries. The impressions, clicks and positions associated with the new response are attributed to that follow-up.
The query and associated landing page can give us clues. They don't prove that a particular row came from an AI conversation or let us reconstruct what came before it.
A query row can be a fragment of a much bigger conversation.
The same goes for images.
Have you ever uploaded a photo to ChatGPT and asked, “What's this?”
I've uploaded photos of my garden and asked what a plant is, how to care for it and how long it takes to grow. Yes, I am a noob gardener.
The point is that I already have the plant. I am not starting with a product search. In the same way, I might already own a car and send ChatGPT a photo to ask about an operational feature.
That interaction starts with a photo. The answer may use information from sources other than the original retailer or manufacturer, without sending me back to either.
AI can help people use something they already own or assess a choice already in front of them.
We don't necessarily know how they found the shop or the product.
Those are different jobs from discovering a brand. In my opinion, they deserve different measures.
A brand appearing in an answer after I've uploaded its packaging tells us very little about whether AI introduced me to it.
I can see value in using real customer questions to test answer accuracy, missing product information and how recommendations change over time.
Those results tell us what happened in the scenarios we tested. By themselves, they don't tell us how often customers have those conversations or how many sales AI triggered.
That's my issue with treating a prompt visibility score as evidence of customer behaviour.
Adding more prompts doesn't give us access to those missing interactions.
I'm comfortable acknowledging that we can't measure all of this. I'm less comfortable with a visibility score being presented as though we can.
There is useful platform reporting, but we need to be precise about what it measures.
Google's Generative AI performance report reports impressions for links to your site in AI Overviews and AI Mode. It lets you examine those impressions by page, country, date and device.
Bing's AI Performance reporting reports citations of your content across supported Microsoft AI experiences. It includes cited pages and a sample of the grounding queries used to retrieve content. Those grounding queries are not necessarily the customer's original words.
These are different measures. An impression is not the same as a citation, and neither proves that an answer changed a purchase.
For the surfaces they cover, these reports provide evidence of content exposure or citation activity. They still don't reveal the full private conversation or every brand mention.
This is where AI search and SEO overlap. We still need traditional organic metrics alongside newer visibility measures, including impressions, clicks, landing-page performance and conversions where we can observe them.
Tracking prompts, especially specific sets of long-tail prompts alone, is not enough to measure AI visibility.
Top comments (0)