In response to this, a new set of tools have appeared on the market to measure 'AI visibility' (how often brands appear in responses from LLMs). While these tools aim to serve a similar purpose to traditional search visibility tools, they face a very different set of constraints: 

​

• Search data: no one has access to actual search volumes across LLMs

• Synthetic prompting: without query logs, tools typically use AI to generate prompt lists that are intended to simulate real user queries, but without a way of validating this

• Probabilistic answers: the same prompt can return a different answer each time - there are no fixed ranking positions

• Model variance: different models (e.g., ChatGPT, Claude, Gemeni) will also generate different responses to the same questions

• Personalisation: results may change depending on user context or chat history

• 'Query fan-out variance': LLM's will address your query by splitting it into multiple sub-queries, combining multiple pieces of information into a single answer

​

From SEO visibility to AI visibility

​

So, what does this mean in practice? Unlike SEO, where visibility is tied to ranking positions and keyword volumes, LLM visibility is more context-dependent. That shift changes the question that matters for advertisers: instead of asking where you rank for certain keywords, the focus becomes how likely your brand is to be mentioned within a given topic or context.

​

At Coppett Hill, we approach this in three steps: 

1. Prompt design: we select a sample of high-intent, high-traffic keywords that drive actual conversions for our clients to build 'common-sense' prompts that reflect real user behaviour and wider search demand. These prompts are skewed towards research-oriented, top-of-funnel queries.

2. Response capture: we record the actual responses generated by LLMs for these prompts, simulating what real users would see.

3. Analysis - the above enables us to focus on a few key areas:

   • Brand mention rate: how often clients and competitors are referenced in generated answers.

   • Source analysis: which sites the LLM draws from, providing insight into which domains the models currently weight as most credible.

   • Citations: whether the LLM explicitly includes or links to a client's (or competitor's) web pages.

​

By segmenting prompts into different topical groups, we help our clients understand the likelihood of their brand being mentioned across the kinds of topics potential customers are searching for, and how that compares to competitors, rather than focussing on visibility as a single-query outcome. As noted in a recent Ahrefs blog: "AI tracking [is] less like rank tracking and more like polling. You don't care about one answer, you care about the direction of the trend across a statistically significant amount of data."

​​

And why does this matter? Because when customers do arrive via AI referrals, they convert well. Across our AI Index sample, AI-attributed traffic converts at 3.9% vs. 2.6% for organic search (excluding one outlier client). This suggests that AI referrals are at least as qualified — and on average, more likely to convert — than users arriving via organic search, reinforcing that optimising for this growing channel can help capture high-intent customers.

​

How are AI Overviews changing search behaviour?​

​

To asses the impact of AI-generated search results, we've measured how often AIOs actually appear in search results for Google (by far the dominant search engine among our clients). Using the same mix of companies as in our AI referral analysis, we tracked the share of advertiser organic traffic exposed to these overviews.