The AI recommended you. Did the shopper choose you?
Key Takeaways
The AI does almost 95% of the talking, the shopper does all the choosing. This dynamic means most brands engaged with and selected were not considered by the shopper before the chat.
Familiarity is powerful, but it is a dormant asset. The shopper rarely brings you into the conversation even if you’re on their list. It only takes effect once the AI has mentioned you.
Most conversations start with a need, not with a brand. To be mentioned, the AI must associate you with the needs of the shopper. Track the needs they bring with them, in their own words, not just the prompts your team writes for them.
Do not judge your AI marketing on mentions. Most brand mentions never capture attention or engagement. Track your warm and cold hook rates, and know which rivals you are winning or losing to in the chat.
This channel does not exist in isolation. Understand your shoppers before and after the conversation, and see where they go next. Then carry the claims that land in the chat onto those pages to reinforce their decision.
A human-shaped hole
Like a chalk outline in a crime scene, there is a human-shaped hole in much of the marketing measurement of LLM conversations. Efforts mainly focus on tracking whether the AI names them, and in some cases they are going a step further to see which brand the AI recommends by the end of the synthetic conversation.
Both measure the machine.
Most AI measurement stops where the shopper starts
| Step | Measured by dashboards |
|---|
There is a chain of influence which gets you into an AI conversation, recommended by the AI, considered by the human, selected by the human, and eventually bought by the human.
It is the human section of that chain which ultimately exerts the most influence on a brand’s outcome
Brands in an LLM conversation follow an attrition curve
Our previous blog post explained why AI visibility and influencing the shopper in the conversation are two entirely different things.
When you follow a brand through a full conversation from first mention to last message, you realise these conversations are more like a game of snakes and ladders than a straight race to the finish.
You also realise these conversations do not exist in isolation. They are one part of a broader path to purchase where external influence determines the shape, purpose and outcome of the conversation.
When it comes to LLM conversations with real shoppers, a brand faces three gates: getting into the chat (or, the “shelf”), being picked up by the shopper, and being selected by the shopper.
To track the conversation and position it within its broader buying context, each of our shoppers went through three steps:
Before the chat. They told us which brands they had already looked at. We call this their starting list, and the AI never saw it. They also gave us their purchase context.
The chat. They prompted and talked to the AI however they liked. We read every message, every reaction, and every search the AI made. We also tracked each brand, from when it first appeared, who said the name first, and what the shopper did with it.
After the chat. They told us what was missing from the conversation that would have helped with their decision, what they would check, where they would go next, and which brands were added or dropped from their list.
Because the AI never sees the shopper’s starting list, we can separate what the AI brought to the conversation from what the shopper brought with them. We follow every brand, on the list and off it, before, during and after the chat.
The AI proposes, the shopper disposes.
Shoppers typically treat the LLM as a utility rather than a peer. They address it with short, blunt statements of need or preference, ignoring much of what the AI throws at them.
“Hi what’s a good laptop? I use spreadsheets keep it under £500”
This transactional dynamic creates an inherent conversational structure that governs the playing field in which brands compete within this channel.
It means the vast majority (229 out of 259) of laptop brands in our study first appear in an AI message. Only 15 of the 75 shopper sessions include a brand brought up by the shopper once the chat began.
It also means the majority of the brands which appear are met with silence. Many are engaged with, some are selected, and only a minority earn the dignity of rejection with a reason.
The AI admits entry, and it can sway a shopper. But it is the shopper who holds the power, and that power is exercised through selective silence and sparse dialogue. Shoppers rarely say no. They answer the one line that interests them and skip the rest. Most brands are part of the rest, and get left behind by the conversation.
The AI does almost all the talking. The shopper makes every choice.
Only shoppers select. Selected means the last maker the shopper chose in the chat, as labelled by a model. It is not a purchase.
| Measure | AI | Shopper |
|---|
Source: Nova Laptop Pilot Study, UK, Aug-Sept 2026. 75 conversations, Claude Sonnet 4.6 with live web search.
This is a familiar dynamic, pushed to an extreme, and stripped of conventional conversation rules. Of the 140,000+ words across the 75 conversations we tracked, almost 95% of them came from the AI. Despite this massive imbalance, it is the shopper who holds all the power. No salesperson who speaks that much will hit their sales target.
Volume beats familiarity.
When one party does nearly all the talking and the other makes every choice, the two forces settle into an odd equilibrium.
Familiarity wins on rate. A brand on a shopper’s starting list has a consistent advantage over those off of the list. When the AI raised a brand on the starting list, the shopper engaged with it in the chat 61% of the time versus 36% for unlisted brands.
The same gap held in the cases where the shopper made a final selection too, where listed brands made the cut in 8 of 49 cases compared to 18 of 180 for unlisted brands.
But the AI wins on sheer volume. Because brand proposals are dominated by the AI, around four times as many unlisted brands showed up across the chats as brands on the list.
“… hadn’t thought about Acer, always used Dell until the assistant gave some good details on it. Now it looks like the best value for my budget.”
This meant unlisted brands had over twice the volume of shoppers engaging with them (64 versus 30), and shoppers chose them as their final selection in the chat in 18 sessions compared with the 12 cases for brands on the starting list.
| Group | Raised | Engaged |
|---|
Source: Nova Laptop Pilot Study, UK, Aug-Sept 2026. 75 conversations, Claude Sonnet 4.6 with live web search.
Most brands do not achieve engagement, let alone selection. Between the first message and the last, the three gates outlined earlier are where brands fall by the wayside. And each of those gates relates to a different live decision.
Gate 1: Getting into the chat
The decision: which prompts should our visibility work track?
Most shopping conversations start as a blank canvas. It is the utility mindset described above in action.
The AI might use natural language, and 3 in 4 shoppers may have used some form of “please” or “thank you”, but ultimately they treat the AI as a tool rather than a peer. They outsource the cognitive heavy lifting right from the start.
In practice, this means shoppers tend to leave the brands considered beforehand at the door, instead focusing almost entirely on buying needs over specific brand names.
“Need a reliable machine, nothing fancy, no gaming, what’s best?”
Only in 12 opening prompts did the shopper name a laptop maker, and only in 5 of those did shoppers want that brand. The rest mentioned it as something they were leaving, repeating, or using as a price reference.
“Want to replace my Dell, I’m not keen on the feel of the keyboard.”
Very rarely did the shopper feel the need to bring a brand on the starting list into the later conversation if it was missing. Of the 102 brand listings across shoppers’ starting lists, 31 of them never came up, and most of the ones which did were mentioned by the AI, not the shopper.
“I started off wanting an Apple, but after explaining what I need it for the AI talked me out of it, so I chose a different laptop”
Last brand selected is the last maker the shopper chose in the chat, as labelled by a model. It is not a purchase.
| Stage | Brands | First raised by AI | By shopper | By both |
|---|
Source: Nova Laptop Pilot Study, UK, Aug-Sept 2026. 75 conversations, Claude Sonnet 4.6 with live web search.
This is the first gate of the conversation, and visibility tools cover this well. They show whether the AI names you, how it describes you, and which sources it draws on. This measurement belongs to GEO, SEO, content and PR teams.
What it cannot currently see is the human side of that moment.
Shoppers orient the conversation around a buying need rather than a named brand. So the prompts that reflect that are the ones real people bring to the conversation. Most brands reach the conversation because the AI matched them to a need, not because the shopper asked for them. Pairing visibility prompt tracking with the shoppers’ buying needs, in their own words, allows you to bring the two in line and identify where the gaps are.
Mental availability still matters in this channel, but it kicks in at the next gate.
Gate 2: Being picked up
The decision: when we’re named but passed over, which fix gets the money: being known, the claims the AI repeats about us, or the proposition itself?
You can achieve AI visibility and still be invisible to the person speaking to the AI.
To understand that problem and do something about it, you need to see what the human does after the AI mentions you.
Similar to social feeds, there is an attention economy inside the conversation. Shoppers selectively engage with only a slice of what the AI presents to them, skipping the rest.
Familiarity can clearly tip the odds in your favour. As we saw earlier, shoppers engaged with brands on their starting list 61% of the time when the AI raised them, against 36% for brands off it.
This gap is not just a reflection of bigger brands being more likely to be on a shopper’s starting list. We compared each brand with itself. The same brand, raised by the same AI under the same conditions, was picked up by the shopper far more often when it was on their starting list. Take Lenovo, for example. Shoppers engaged with it in 13 of the 18 instances where it was on their list, and only in 9 of 27 times when it was not on the list.
So the structure of an AI shopping conversation produces somewhat of a blank canvas filled in by the AI. That canvas is not immune to the influence of brand familiarity. A smaller brand might increase its chances for visibility and optimise the claims or phrasing the AI uses to describe it, but it still runs into the human attention filter which results in so many brand mentions being skipped.
AI visibility is not a substitute for brand building, because familiarity travels into the conversation.
At the same time, the volume of AI proposals works in the opposite direction. The AI proposes far more unlisted brands than listed ones, which can favour smaller brands and threaten bigger ones. Familiarity wins on rate, volume wins on scale.
This is the warm and cold hook rate: how often shoppers engage with your brand when they already know it, and when they don’t. Hook rates often hinge on specific claims the AI makes about your brand which the shopper picks up.
“Tell me about option 3, I like 1TB of storage and the OLED screen.”
Similar to a brand tracker measuring awareness and consideration, in the age when consideration forms and falls apart across billions of AI conversations, warm and cold hook rates give us different problems with different budgets:
You lose cold mentions but hold warm ones. Shoppers don’t know you. That is a job for brand media.
You lose warm mentions too, then you need to look at what the AI says about you. Nova shows which claims shoppers stop for and which they pass over. That points content and PR at the claims that land with people which is relevant to all channels.
Shoppers pick up the rival named beside you, then your proposition does not stand out in a list of five options. This is a product marketing problem.
Gate 3: Being selected
The decision: What should our AI programme be judged on?
Most AI marketing efforts are currently judged on mentions. The third gate shows us why that approach is not sufficient by itself.
Thirty-two of the 75 sessions ended with the shopper selecting a brand. Twelve went to a brand on the shopper’s starting list, twenty to a brand off the list. In both cases, most of those selected brands first appeared in an AI message, not a shopper one.
You can judge the AI’s influence on your marketing efforts in two ways:
Winning cases: how often the shopper who did not have you in mind chose you.
Losing cases: how often shoppers who did have you in mind selected a rival, and which rival.
The first reflects growth in the AI channel, and the second shows you leakage of intent and who captured it in your stead.
Selection in an AI conversation is the front-runner once the chat ends, not necessarily the brand winning the purchase. Shoppers’ circumstances might change, and subsequent channels used along the journey can still reshape the decision. But inside the conversation, selection is the clearest signal of who won the exchange.
“I know which one I’ll probably get, need to do a bit more research before the final decision”
After the chat: the handover
The decision: Which pages need to back up what the AI says about us?
The chat is not the end. Asked what would have made it an easy decision, 31 of 75 said seeing reviews or ratings, and 30 said proof that the AI’s claims were accurate. Only 18 told us nothing was missing from the conversation to help their decision.
People are not always as meticulous as they make out, and they seem to accept the risk of AI inaccuracy to some extent. Only 20 of them planned to check something specific from the conversation elsewhere.
“It told me to double check the specs myself, which made me wonder how far to rely on it.”
Nevertheless, the majority (69 of 75) still planned to go elsewhere.
Google came first (33), then retailer sites (28), the brand’s own site (27) and review sites (26).
Seeing inside the conversation will let you understand which claims land with the shopper, so you can carry these claims over to where the shopper goes next in the journey to reinforce your place in their consideration:
List the claims that made an impact on engagement or selection.
Check them against the next stop in the journey, and segment this by where shoppers are in their journey when they enter the conversation versus where they intend to go afterwards. Open your retailer listings and product pages to see if they lead with those claims with supporting reviews alongside them.
Fill the gaps your shoppers name. In this study, reviews and ratings were the biggest gaps which would have made the decision easier.
What to measure at each gate, and who owns the budget
| Gate | What to measure | Already available elsewhere | What only shopper sessions add | Budget owner |
|---|
Most of the industry is still making sense of AI’s impact on marketing through the lens of web traffic. That view starts at the end of the chain and works backwards. The three gates invert it. They start with the shopper; the needs they bring, the brands they already carry, the ones they pick up or pass over, the one they select, and the stops they plan to make next.
When you measure from the human side, the conversation is not just a source of visits. It is where consideration is won and lost, before anyone clicks.