What 10,891 AI Answers Showed About A Retailer's 41.9% Share Of Voice

The Problem

AI assistants were sending this retailer more visits each year while organic sessions were slipping, so we needed to know how it really showed up in AI answers and what to change. A share of voice score on a dashboard can't tell you that, because it doesn't say which products you're named for, where the AI got its information or whether the prompts behind the score are any good.

What I Did

• Ran an AI audit in July, with 24 buyer prompts on four AI platforms, twice each, and 13 checks on the sources AI tools cite, the listings and reviews they lean on and how they describe the brand.
• Exported every answer from Prompt Watch, 17,998 from 21 July to 27 September 2026, and broke them down by product group, model and source instead of relying on the dashboard score.
• Checked 26 key pages against what the answers were saying, then turned that into 16 plain actions for the client and its developer. They covered wrong or conflicting product details, delivery promises that contradicted each other and review scores AI tools couldn't read.
• Briefed the main category pages from October so they help someone choose, with comparison tables across the ranges, because AI tools were citing rivals' buying guides where the retailer's category pages were too generic.

What I Found

The retailer was the most-named brand of any kind in these answers, ahead of the big hardware makers, and it appeared in 2,369 of the 10,891 answers. That is a good headline, but the detail underneath it is what told us what to do.

Being Named Is Not The Same As Being Cited

Of those 2,369 answers, 1,074 named the retailer without linking to its site. The gap changes a lot by product, so server racks were named in 48% of answers and cited in 32%, while ethernet cables were named in 22% and cited in just 5%. Network switches and UPS barely registered at all, which the headline score hides.

One Answer Proves Very Little

I looked at the 331 prompt and model pairs with at least 20 runs. In 133 of them the retailer was never named, in none of them was it named every time, and in 139 of them, or 42%, it was named somewhere between 10% and 90% of the time. So a screenshot of one answer tells you very little, and you need repeated runs before you can say anything.

Some Prompts Were Measuring The Wrong Thing

On the "wall mount cabinets" prompts, 1,214 of the 1,910 answers mentioned kitchens, garages or home stores, so that product group's score meant very little and I flagged the wording. Prompts naming a city also named the retailer 33.9% of the time against 18.4% without one, so the mix of prompts moves the score nearly as much as anything on the site.

AI Leans On Sources You Don't Own

Reddit was cited in 2,560 of the 15,898 answers and YouTube in 2,028, against 2,208 for the retailer's own site. On reliability questions, two review sites were cited in 589 and 379 of 761 answers, against 214 for the retailer's own testimonials page, so review management is now part of how a brand gets described in AI answers.

What Happened

Visits from AI assistants rose from 2,122 to 4,728, comparing January to September 2025 with the same months of 2026, while organic sessions fell from 305,587 to 278,081. AI is still a small share of visits, but it is the part that's growing.

How I Measured It

Share of voice is the retailer's share of all mentions of it and its tracked rivals, on non-branded prompts only. Our tracking data only starts in July 2026, so there's no earlier baseline and the 41.9% is a snapshot rather than a rise.

AI assistants send more visits to most sites each month, so I don't credit the rise in AI visits to my work alone. The audit actions and the category page rewrites are also too recent to show up in these numbers, so I'm not claiming them.

Gemini's citations arrive as redirect links, so they all look like google.com, and at first that made Gemini look like a gap. Once I resolved 540 of the 549 links in the July audit, Gemini turned out to be the best platform for citing the retailer's own sites, so I corrected that finding before it went anywhere.

What I Would Do Next

I'd run the same prompts again once the rewritten category pages have been live for eight weeks and compare named and cited rates by product group, because that's the first real before and after this work can have. I'd also rewrite the prompts that drift off topic, so the score measures the products the retailer actually sells.