London Cookie and Bakery AI Visibility Index 2026: which dessert brands ChatGPT, Gemini and Google AI name on their own

Last updated: 8 September 2026

Most categories we measure are flat: no brand comes close to owning the answer. London cookies, bakeries and desserts is the exception, and that makes it a more useful test case, not a less interesting one. On 8 September 2026 we put 200 real diner questions to ChatGPT, Gemini and Google AI Overview across London, and recorded 1,449 answers. We tracked 31 cookie, bakery and dessert brands, and 23 of them came up at least once. Every brand was scored against that same 1,449-answer denominator. One brand leads outright, and the finding worth remembering is where that lead comes apart.

The numbers in brief

• Puffy Cookies is named most often, in 20.01 percent of the 1,449 answers, well ahead of The After School Cookie Club on 12.63 percent.
• The median across the 23 brands that appeared at least once is 1.45 percent.
• Eight of the 31 brands we track did not appear in a single answer.
• On the visitor and first-trip discovery question, the same leader falls to 2.27 percent.
• thatsup.co.uk is the most-retrieved source, appearing in 11.94 percent of answers.

What we measured, and how

We track 200 questions phrased the way a diner actually types them. They span eight themes: which cookie shop is simply the best, where to get dessert in London more broadly, quality and provenance, occasions and gifting, neighbourhood-specific searches, quick and casual value, dietary and style needs, and first-trip visitor discovery. Every question is anchored to London, or to a landmark inside it, so a model answers about this city and not a different one.

The figures in this report are drawn from 1,449 answers recorded on 8 September 2026 across ChatGPT, Gemini and Google AI Overview. We record what the engine returns, not what it was asked to return. Thirteen of the 200 tracked questions name a brand directly, or ask for a head-to-head against a named rival; we exclude those from every figure here, because they measure whether a model can repeat a name we handed it, not what it recommends unprompted.

One honest note on the leading figure. Every brand in this report is scored against the same 1,449-answer denominator, including the leader. A single answer inside the quick-and-casual-value question set did not register a scored value for the leading brand specifically, while every other brand's row was unaffected. We use the full 1,449-answer denominator for every brand's percentage here, the leader included, so the 20.01 percent figure above already reflects that.

Three measures, in plain words:

• How often AI names them: the share of the 1,449 answers in which the brand is named.
• Share of the mentions: the brand's portion of all 1,910 brand mentions recorded across every answer.
• Average position: where the brand tends to land in a list when it is named. Lower is better.

No brand paid to be included, and the names in this report identify research subjects, not customers and not endorsements.

Who the engines actually name

The gap at the top of this table is real. Puffy Cookies leads the next name by more than 7 percentage points, a wider margin than most categories we track show between first and second place. But the field falls away fast beneath the top few: the median across the 23 brands that appeared at least once is 1.45 percent, and outside the published fifteen no brand reaches even 0.6 percent, with Toad Bakery the next name down at 0.55 percent. Eight of the 31 brands we track did not appear in a single answer.

The finding that matters more than the ranking

A firm number one on the everyday question conceals a sharp cliff at the edges of the category. Puffy Cookies leads outright on the question that most closely names its own product, best cookies in London, at 62.56 percent of that topic's 211 answers, and on quality and provenance at 47.41 percent of 116. It also leads the neighbourhood-specific question at 18.29 percent of 175, and the quick and casual value question at 16.18 percent of 173.

Move one step wider, though, and the picture changes fast. On "where to get dessert in London," a broader question where a cookie specialist has to compete against bakeries, patisseries and doughnut shops rather than only other cookie brands, Puffy Cookies is not even the name the engines default to: that topic's 212 answers go to Crosstown, named in 9.43 percent of them. On occasions and gifting, Crumbs and Doilies leads at 17.74 percent of 186. On dietary and style questions, The After School Cookie Club leads at 35.26 percent of 156. And on the specific question a first-time visitor asks, what to try on a trip to London, Puffy Cookies falls to 2.27 percent of 220 answers, a fraction of its 20.01 percent overall share and level with names far lower down the main table.

The lesson generalises beyond this one brand. Winning the question closest to your own product does not automatically carry over into the wider questions around it. A specialist identity raises the ceiling on the question it was built to answer and lowers it everywhere else.

Named often and positioned well are the same story here

Puffy Cookies leads on both measures that usually pull apart at the top of a leaderboard. It is named in more answers than any other brand, and when it is named, it lands closer to the top of the list too, at an average position of 1.9, the best average position of any brand in the tracked field. The next best average position among the top eight is 2.6, held by Blondies Kitchen, and most of the rest trail well behind that. That combination, named often and placed well, is what makes the category-question lead so strong, and it is exactly why the drop-off on the broader dessert and visitor questions is worth taking seriously rather than dismissing as noise: the mechanism that wins the narrow question does not transfer to the wide one just because the brand is otherwise strong.

Where the answers come from

Three things stand out.

First, no single editorial title dominates the way it does in other London food categories. The top slot goes to thatsup.co.uk, a listings aggregator, at 11.94 percent, with The Infatuation tied for the next-highest spot alongside Instagram itself, both at 10.14 percent.

Second, visual platforms carry unusual weight here. Instagram is retrieved as often as The Infatuation, and a rival brand's own domain, cremelondon.com, is retrieved in 8.14 percent of answers, more than Time Out's 6.42 percent. A cookie or bakery brand's own photography and product pages are doing real work in this category, not just its press coverage.

Third, community sources outrank some editorial ones outright. Reddit, at 5.38 percent, is retrieved more often than Time Out. The operating instruction for anyone in this category: your own site's product pages and imagery matter here more than in most food categories, so treat them as a primary channel, not an afterthought behind press coverage.

What this report can and cannot tell you

These results are directional, not definitive. They describe one question set, on three engines, on one day.

Read this way, the data can answer a bounded set of questions:

• Which London cookie, bakery and dessert brands the engines name without being prompted with a brand name.
• How concentrated the field is on a category-specific question, and how that concentration disappears on the wider questions around it.
• Which source types the engines reach for when they build a London cookie or dessert recommendation.
• How a brand's naming frequency compares with how much room it takes up inside an answer it appears in.

It cannot do several other things, and it is worth being blunt about them. It cannot prove that appearing more often produces more footfall or more revenue. It cannot establish a causal link between any single change a brand made and its position here. And it cannot tell any individual bakery that it should appear in every one of the 200 questions, because most brands have no real business appearing in most of them. A single-product cookie specialist has no natural claim on a full patisserie's wedding-cake question.

The operating instruction is therefore not to chase the leaderboard as one blended score. Split the questions by the ones a brand could genuinely win, read the exact answers the engines give for those, and repair the weakest layer of evidence underneath them.

What this means if you run a cookie, bakery or dessert brand in London

If you run a single-product or niche specialist brand, the topic breakdown above is the more useful map than the single overall rank. Winning the specific product question your brand is built around, the way Puffy Cookies wins "best cookies in London," does not automatically carry over into the wider dessert or visitor-discovery questions, where a different set of names leads outright. Read the exact answers for the topic closest to what you actually sell before assuming a strong niche position has become a broad one. Our companion piece on London's burger scene shows the opposite shape worth comparing against: a genuine two-name cluster at the top, rather than one brand far ahead everywhere except at the edges of the category.

If you run more than one site, the same principle we found in the burger index holds here too: a well-documented individual location tends to pick up more room inside an answer than a single flagship page speaking for the whole group. Check each site as its own entity rather than leaning on one strong location to carry the brand.

If you're weighing a niche identity against a broader offer, the visitor-discovery cliff in this data is the concrete illustration of the trade-off. Specialising raises your ceiling on the question closest to your specialism and lowers it everywhere further from it. For a wider view of how much flatter most categories are by comparison, our UK Restaurant AI Visibility Index and Texas Dining AI Index both track dining categories where no single brand gets anywhere near this level of dominance on any one question. And the same measurement holds up well outside hospitality: our indexes for UK tech recruitment, contact centre QA software, crypto wallets and Middle East on-demand strategy consulting all show the same lesson: a strong position on your core question rarely extends to the adjacent ones for free.

Sources

• Schmitdy category tracking, London cookie, bakery and dessert question set, 8 September 2026. The 1,449 answers, 200 questions and 31 tracked brands behind every figure in this report.
• Puffy Cookies, puffycookies.co.uk. Official site of the most-named brand in the index.
• The After School Cookie Club, theafterschoolcookieclub.com. Official site, second in the published table.
• Crumbs and Doilies, crumbsanddoilies.co.uk. Official site, third in the published table.
• Ben's Cookies, benscookies.com. Official site, fourth in the published table.
• Floozie Cookies, flooziecookies.com. Official site, fifth in the published table.
• Crosstown, crosstown.co.uk. Official site, sixth in the published table.
• Blondies Kitchen, blondieskitchen.co.uk. Official site, seventh in the published table.
• Cutter and Squidge, cutterandsquidge.com. Official site, eighth in the published table.
• Ole and Steen, oleandsteen.co.uk. Official site, ninth in the published table.
• Creme London, cremelondon.com. Official site, tenth in the published table.
• Fortitude Bakehouse, fortitudebakehouse.com. Official site, eleventh in the published table.
• Arome Bakery, aromebakery.co.uk. Official site, twelfth in the published table.
• Bread Ahead, breadahead.com. Official site, thirteenth in the published table.
• E5 Bakehouse, e5bakehouse.com. Official site, fourteenth in the published table.
• Dominique Ansel, dominiqueansellondon.com. Official site, fifteenth in the published table.
• thatsup, thatsup.co.uk. The most-retrieved source across the tracked answers.
• The Infatuation, theinfatuation.com. Tied for the highest-retrieved editorial title.
• Instagram, instagram.com. Most-retrieved social platform.
• Reddit, reddit.com. Most-retrieved community source.

Every brand named in this report is a research subject. Inclusion is not an endorsement, and no brand paid to appear.

If you want to know which of these questions your own brand already shows up in, we can run the same measurement for your market.