Generative AI in Local Search: What a Pizza Case Study Revealed
In July 2023, a team based in Brighton ran the same five local pizza searches through four generative AI tools and two traditional search engines.
The aim was simple: see whether the answers were accurate and useful for someone trying to find a local business. This page explains what that snapshot showed, how the tools differed, and what the findings mean for local visibility work today.
How the test was set up
The test treated the searcher as a typical consumer looking for pizza in Brighton. Five queries were chosen to cover slightly different intents, all centred on local business discovery: best pizza restaurants, top-rated pizza restaurants, most authentic pizza restaurants, best takeaway pizza, and pizza delivery near me.
Those exact queries went into four publicly accessible generative tools: Google Bard, Search Generative Experience (SGE), Bing Chat, and OpenAI’s ChatGPT (May 24 version). Two traditional engines, Bing and Google, acted as a control group. Screenshots were taken of every result so the team could review the type of content and media shown, whether sources were quoted, and how accurate the information was. Prompts were not refined and the bots were not asked follow-up questions about where their information came from.
One important qualification: this is a snapshot of tools as they stood in July 2023. The technology has changed considerably since then, so treat the figures as a dated reference rather than a description of current behaviour.
What the snapshot found
The headline results were mixed. Traditional search engines remained the most accurate source of business information. SGE showed local business information — listings, reviews and maps — in every result, compared with 80% for traditional Google. Bing showed directory links, maps, images, review ratings and business listings in every result.
Bard was the weakest performer on accuracy: 80% of its results contained incorrect information, such as wrong business names or restaurants located elsewhere in the UK. Bard and ChatGPT generally did not provide citations to support their answers. Bing Chat, by contrast, cited its sources in every result.
The table below records how often each media format and type of business information appeared across the five queries. It is the original July 2023 data and has not been re-run.
Media formats and business information by tool
The table below shows, for each tool and engine, the percentage of the five queries where a given format or information type appeared. Read it as a format-presence map, not as a quality score: a tool can show a map every time and still name the wrong restaurant. The protected data module table0001 is reproduced in full alongside this page.
What each query revealed
Best pizza restaurants in Brighton. Bard’s list was a grab-bag: independents, familiar chains, shopping-centre food-court names, and a London pizzeria that has nothing to do with Brighton. No review ratings appeared, and nothing explained why these counted as the ‘best’. SGE produced a local-pack layout with a map and ratings, but the one-line descriptions were so generic they could have described almost any casual dining room. Bing Chat gave a short list with clickable sources, though one source was misattributed. ChatGPT named mostly well-known Brighton independents, but included a steakhouse that has never served pizza and described a Neapolitan pizzeria as New York-style.
Top-rated pizza restaurants in Brighton. Bard served up the same set, London entry and all, and added Google ratings that did not match the genuinely top-rated restaurants. SGE again used a map-pack layout, but its pizza-style tags were wrong: one restaurant was tagged sourdough despite not making sourdough, and a national chain was tagged Neapolitan. Bing Chat drew on TripAdvisor reviews but omitted star counts and review volumes, and its order differed from TripAdvisor’s own ranking. ChatGPT declined to supply current information and pointed only to TripAdvisor, Yelp and Google Maps.
Most authentic pizza restaurants in Brighton. Authenticity is a matter of definition, so accuracy here depends on the reader. Bard listed five Brighton independents with largely accurate descriptions but no stated reasoning. SGE’s basis was unclear and its list included a tourist buffet restaurant. Bing Chat quoted a well-known local Italian food writer’s guide, a strong signal that it had read the intent. ChatGPT’s top pick was a restaurant that does not serve pizza, though it added a fair caveat that tastes vary.
Best takeaway pizza in Brighton. Bard recycled the same restaurants without any takeaway-specific detail such as delivery or courier availability, and this time included two London restaurants. SGE labelled results as pizza delivery services, but only two of the three offered online ordering. Bing Chat produced a muddled list with duplicates and a business name that did not exist, apparently blending two local brands. ChatGPT gave mostly accurate independent recommendations, apart from the recurring New York-style error.
Pizza delivery near me. Bard again included two London restaurants even with location settings on. SGE gave its most useful results for this query, listing outlets built around delivery and visible on the main food-ordering apps, though the descriptions sat oddly against average ratings. Bing Chat returned a thin result. ChatGPT pointed to delivery platforms and explained how to use them and a search engine.
How the tools compared
ChatGPT and Bard were rated very low for usefulness on local search queries, and both produced the highest share of inaccurate information. That said, few consumers were likely to use either tool for local discovery at the time; both position themselves as collaborators for tasks such as coding or menu planning. A dentsu survey cited in the original study found that 73% of 1,000 US adults had not used generative AI tools, and those who had mostly cited experimentation or content generation.
Bing Chat and SGE were a closer call. Because they sit inside traditional search engines, their results felt more familiar and were generally accurate, with a few rough edges: SGE’s business labels were sometimes wrong or irrelevant, and Bing Chat’s cited sources did not always match the information shown. Traditional Bing also performed strongly on intent matching, showing maps, images, ratings and listings in every result compared with 80% for Google.
A practical checklist for local businesses
Use this checklist to decide how much weight to give AI answers in your local visibility work.
- Separate presence from accuracy. A tool that shows a map, rating and listing every time can still name the wrong business or the wrong cuisine. Check the factual claims, not just the format.
- Check location handling. In this snapshot, Bard repeatedly returned London restaurants for Brighton queries, including for ‘near me’. If location matters to your category, verify that the tool is actually using it.
- Look for citations. Bing Chat cited sources in every result; Bard and ChatGPT generally did not. A cited answer is easier to audit and easier to influence.
- Treat labels as claims. SGE’s pizza-style labels were sometimes wrong. If a tool describes your business, check whether the description is accurate and whether you can correct the underlying data.
- Keep the fundamentals strong. Accurate business listings, a maintained Google Business Profile, a healthy review profile across relevant sites, and community engagement all support the likelihood of being cited.
- Do not ignore non-Google engines. Bing’s strong showing on intent and format is a reminder that visibility beyond Google can reach different audiences.
A worked example: the VIP mislabel
Across several queries, ChatGPT described the Brighton restaurant VIP as ‘known for its New York-style pizza’. In reality, VIP is a Neapolitan pizzeria. The error is small in wording but meaningful in practice: a reader looking for New York-style pizza might choose VIP for the wrong reason, and a reader looking for Neapolitan pizza might skip it. The same pattern appeared with a steakhouse that has never served pizza and with a restaurant that does not offer pizza at all. The lesson is not that AI is always wrong, but that a confident-sounding description can be factually wrong in a way that affects a real decision. For a local business, the practical response is to make accurate, consistent information available across the sources these tools draw on.
What this means now
The July 2023 snapshot showed that standalone chatbots could not beat traditional search for local business discovery, while AI features embedded in search engines produced results that felt familiar and were mostly accurate. The lines were blurry then, and the tools have changed a great deal since. The study’s own recommendations point the same way: accurate listings, a maintained profile, reviews across relevant sites, and engagement all support being cited by AI tools as well as by traditional engines. Treat the figures on this page as a dated reference and check current behaviour before making decisions.
| Format / information | Bard | Bing Chat | Bing Search | ChatGPT | Google Search | SGE |
|---|---|---|---|---|---|---|
| Website links | 100% | 80% | 80% | 0% | 100% | 40% |
| Directory links | 100% | 80% | 100% | 0% | 100% | 60% |
| Map | 0% | 60% | 100% | 0% | 80% | 100% |
| Images | 100% | 60% | 100% | 0% | 80% | 100% |
| Review ratings | 20% | 60% | 100% | 0% | 80% | 100% |
| Business listings | 0% | 60% | 100% | 0% | 80% | 100% |
| Sponsored content | 0% | 0% | 60% | 0% | 20% | 0% |
| Inaccuracies | 80% | 20% | 0% | 60% | 0% | 0% |
Original July 2023 data; not re-run. Percentages are the share of five queries where the format or information type appeared.
Frequently asked questions
Which generative AI tool was most accurate for local search in this study?
In the July 2023 snapshot, Bing Chat and SGE were the most accurate of the generative tools, with Bing Chat citing sources in every result. Bard had inaccuracies in 80% of results and ChatGPT in 60%. Traditional search engines remained the most accurate overall.
Why did Bard show restaurants in London for Brighton queries?
The study does not establish the cause. What it records is that Bard returned London restaurants for Brighton queries, including for ‘pizza delivery near me’ with location settings on. That is a location-handling failure in the snapshot, not an explanation of the underlying mechanism.
Does this mean AI tools are useless for local discovery?
No. It means that in July 2023 standalone chatbots were weak for local discovery, while AI features embedded in search engines were closer to traditional search in accuracy and familiarity. The tools have changed since, so the practical step is to check current behaviour for your own category and location.
What can a local business do to be cited by AI tools?
Keep business information accurate and consistent across relevant listings sites, maintain your Google Business Profile and review profile, run review campaigns across multiple relevant sites, and build reputation through customer service and community engagement. These steps support traditional and AI visibility alike.
| Bard | Bing Chat | Bing Search | ChatGPT | Google Search | SGE | |
|---|---|---|---|---|---|---|
| Website links | 100% | 80% | 80% | 0% | 100% | 40% |
| Directory links | 100% | 80% | 100% | 0% | 100% | 60% |
| Map | 0% | 60% | 100% | 0% | 80% | 100% |
| Images | 100% | 60% | 100% | 0% | 80% | 100% |
| Review ratings | 20% | 60% | 100% | 0% | 80% | 100% |
| Business listings | 0% | 60% | 100% | 0% | 80% | 100% |
| Sponsored content | 0% | 0% | 60% | 0% | 20% | 0% |
| Inaccuracies | 80% | 20% | 0% | 60% | 0% | 0% |