The Bakery ChatGPT Forgot: A Small Business Owner's Wake-Up Call on AI Search

Priya Shah has run The Bristol Bake House on Gloucester Road for eleven years. Locals know her sourdough, the queue on Saturday mornings, the fact that she still writes the specials board by hand. What she did not know, until her nephew asked ChatGPT where to get good bread in Bristol, was that the assistant recommended two other bakeries and never mentioned her at all.

That small, slightly wounding moment is becoming common. Search is no longer just a list of blue links you can rank on with the right keywords. On 14 May 2024, Google announced it was rolling AI Overviews out to Search in the US, an AI generated summary that sits above the traditional results and answers the question directly (Liz Reid, "Generative AI in Search: Let Google do the searching for you", 14 May 2024, blog.google). Less than a year later, Google pushed further, expanding AI Mode, a fully conversational search experience, to more users in the US (Robby Stein, "AI Mode: expanding what's possible with Search", 5 March 2025, blog.google). Perplexity, Gemini and ChatGPT have been doing something similar for longer: instead of ten results, the user gets one paragraph, and only a handful of businesses get named in it.

For a business like Priya's, that is the whole game. If the AI names two competitors and not you, you have effectively vanished from that customer's decision, no matter how good your website's meta description is.

Why some businesses get named and others don't

Generative engines are not using the same playbook as classic Google search. Researchers at Princeton and Georgia Tech studied this directly and published one of the first formal frameworks for it, calling the practice Generative Engine Optimization, or GEO. Testing a range of content changes across thousands of queries, they found that fairly simple edits, adding citations, quoting statistics, including relevant terminology, could increase how often a source was cited in an AI generated answer by as much as 40 percent (Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, "GEO: Generative Engine Optimization", arXiv, 8 November 2023, arxiv.org/abs/2311.09735).

In plain terms, these systems reward content that reads like something worth quoting. They are pulling from pages that state facts plainly, back them up, and are structured in a way a language model can lift a clean sentence from. A beautifully designed homepage with a hero image and the word "artisan" in cursive gives a model almost nothing to quote.

The other ingredient is trust signals that were already familiar to local businesses, just weighted differently now. Reviews, consistent business listings and clear, current information about hours, location and what you actually sell all feed into how confidently an AI system will name you. BrightLocal's long running Local Consumer Review Survey has tracked for years how heavily people lean on reviews before choosing a local business, and that same review data is exactly the kind of structured, trustworthy signal generative engines draw on when deciding who to recommend (BrightLocal, "Local Consumer Review Survey 2024", brightlocal.com/research/local-consumer-review-survey).

What Priya actually changed

Once she understood the problem, Priya did not redesign her website. She rewrote it. Her About page went from a paragraph of brand voice to a set of plain facts: founded 2013, sourdough baked fresh from 6am, gluten free options available Tuesdays and Fridays, address and opening hours repeated in text rather than buried in an image. She asked her regular customers to leave Google reviews that mentioned specific things, the cinnamon buns, the coffee, the dog friendly courtyard, because specific reviews give language models specific things to quote. She also made sure her business listing, her website and her Instagram all stated her hours and location identically, since inconsistency is exactly the kind of thing that makes an AI system hedge and leave a business out.

None of this was expensive. It took her and a part time assistant about two weeks. Three months later, her nephew ran the same ChatGPT query again. This time, The Bristol Bake House was the first name mentioned.

The takeaway for small businesses

Generative engines are becoming a genuine referral channel, sitting somewhere between search and word of mouth. They reward clarity, specificity and evidence over polish. A business that states plainly what it does, where it is and why people like it, backed by real reviews, is simply easier for an AI model to trust and repeat.

That is a fixable problem, not a mysterious one. It does mean treating your online presence as something a machine needs to read and quote correctly, not just something a human needs to enjoy looking at.

If you want to see how your own business currently shows up when people ask ChatGPT, Perplexity or Google's AI Overviews about what you do, that is exactly what we built Ozvor to track. Have a look at ozvor.com.