Ask AI for a local business twice and the list changes. Count how often you are on it.

Several studies from 2026 asked the same question again and compared the lists. The names change from run to run, so one screenshot of one answer tells you little. This post turns the studies into a count you can run on your own business, with a range that shows how much to trust it.

What five studies found when they asked again

BrightLocal published its Local AI Visibility Study on September 16, 2026. It ran about 200,000 non-branded local prompts on ChatGPT, Google AI Mode and Google AI Overviews, repeated them across a map grid, and collected about 1.9 million citations. Two result sets for the same prompt overlapped by only 21% to 33% on average. A business that appeared once showed up again in about half of the repeats. INDEPENDENT RESEARCH

Steady Demand ran one design twice, in July and again in September 2026, on Gemini and ChatGPT. It covered the 50 largest US metros and 10 service verticals, including plumbers, roofers, HVAC contractors and electricians. When it repeated the same question, Gemini named the same top business 7.9% of the time in July and 5.1% in September. Google’s local pack returned the same top listing 90.2% and 83.3% of the time. INDEPENDENT RESEARCH

SparkToro and Gumshoe.ai tested product prompts in late 2025. Volunteers ran 12 prompts on ChatGPT, Claude and Google’s AI 60 to 100 times each, for 2,961 runs in all. The chance of getting the same list of brands twice was under 1 in 100. Strong brands still came back often: the top headphone brands appeared in 55% to 77% of 994 responses. Those are product categories with many competitors. BrightLocal notes that a local pool is usually much smaller, and found less variance for local prompts than for general ones, though still a lot. INDEPENDENT RESEARCH

Two smaller tests point the same way. Custom Legal Marketing asked ChatGPT, Claude and Google AI Mode for law firms in 100 US metros. Of 19,692 firms named, 578 (2.9%) appeared on all three assistants, and the three named the same top firm in 0.9% of 3,200 prompts. INDEPENDENT RESEARCH

A Bakersfield agency, Clicks Dynasty, compared 20 searches on ChatGPT and Perplexity and found that 24 of the 99 businesses Perplexity listed (24%) also appeared in ChatGPT’s top five. COMMUNITY DATA

Ask twice and you may get two lists. Judge your visibility by how often you appear, not by one answer.

What the studies do not show

Where the picks come from

BrightLocal also counted where the answers pointed. Business websites were 93% of the unique domains cited and 42% of all 1,967,762 citations. Google Business Profile, a single source, was 28.5%. INDEPENDENT RESEARCH

BrightLocal says Google Business Profile matters most on AI Mode and AI Overviews and is “hardly used” by ChatGPT, which leans on Yelp and Bing. ChatGPT’s picks overlapped with Google AI Overviews by 20% and with AI Mode by 19%. INDEPENDENT RESEARCH

Steady Demand’s tables show a similar split and how quickly it can change. INDEPENDENT RESEARCH

DHM INTERPRETATION For a service business, Gemini and Google’s AI lean on your own website and your Business Profile. ChatGPT leans on directories your website does not control. That makes two jobs: make your website the best answer to what a customer asks, and keep the directory facts that ChatGPT reads accurate. Neither study says how to win a place in an answer, and the shares moved a lot in six weeks.

BrightLocal also compared businesses with a public storefront and service-area businesses, which have no customer-facing address. The average mention rate was 33.6% for storefronts and 29.4% for service-area businesses. BrightLocal says a fixed public address makes a recommendation much more likely. INDEPENDENT RESEARCH

DHM INTERPRETATION The difference is a few points. A service-area business should state its service area, hours and licensing in plain text on its own pages. It has no storefront address for an assistant to use as a reference point.

Google’s October 15 change points at your own website

On October 15, 2026, Google announced that AI Mode and AI Overviews show more inline links, and show them higher in responses more often. Google also said it will test a display for one or two official pages. They would sit in a dedicated section at the top of an AI response, with context about why each link is useful. GOOGLE DOCUMENTED The post does not mention local businesses, and the official-pages display is a test.

DHM INTERPRETATION If the test reaches local queries, the page Google selects for your business could receive the click. That page should explain who you are, where you work and what the job involves. Business websites were also the largest source type in BrightLocal’s data. Bing reports the other side: its AI Performance report lists which of your pages its AI answers cite, as we explained in an earlier Sauce post.

Check your own mention rate

A mention rate is the share of your checks in which an assistant named your business. A check is one question asked once in a fresh chat. Ask 5 questions 6 times each and you have 30 checks. Count how many of them named your business. The calculator turns the count into a rate and a range, because a rate from a few checks can be far off. The starting numbers are examples. Replace them with yours.

DHM INTERPRETATION With 10 mentions in 30 checks, your rate is 33%, with a 95% range of 19% to 51%. With 20 in 60 it is still 33% and the range narrows to 23% to 46%. With 33 in 100 it narrows to 25% to 43%. That is why a move from 20% to 30% between two small tests tells you little. The range is a standard Wilson interval and assumes each check is independent. Repeated runs of one question may not meet that assumption.

SparkToro says brand visibility needs at least 60 to 100 runs of a prompt. The repeat results above show why. A small business cannot run 100 checks by hand for every question, so pool several questions and keep the same set from month to month. A rate you can repeat is more useful than a precise rate you cannot. INDEPENDENT RESEARCH

Do this week: set up a repeatable check

1. Write five questions a customer would type. Steady Demand used three phrasings for every town: “best plumber in Surrey”, “top rated plumber near Surrey” and “who is a good plumber in Surrey”. Add a question about price and one about a nearby town. Our Ask Maps test has ten prompts you can use.
2. Ask each question six times in fresh chats. Use ChatGPT, Google AI Mode and Perplexity. When the tool allows it, use a signed-out or private window. Record yes or no for whether your business was named, and note the sources shown under the answer.
3. Enter the counts in the calculator and save the result with the date. Keep the same five questions next month so the two rates can be compared.
4. Read the sources the answers show for your competitors. If they are directories you are missing from, claim those listings. If they are business websites, compare their service pages with yours.
5. Make one page answer the customer. Name the service, the towns you work in, your hours, your licence or insurance, and what affects the price, in plain text. Match the details to your Business Profile.
6. Read the queries behind your impressions. Bing Webmaster Tools and Google Search Console both list them. BrightLocal suggests using them to find the conversational questions your site should answer, and gives this Search Console filter for queries of ten words or more: Performance, Query, Custom (regex), ^(?:\S+\s+){9,}\S+$.

What our own numbers show

We read our own data on October 21, 2026. Google Analytics 4 showed 137 sessions between July 22 and October 19, with none from an AI assistant source. Bing Webmaster Tools showed 192 impressions and no clicks between August 31 and October 18. It listed 387 of our pages in its index, up from 58 on September 1. DHM DATA

Many of the queries Bing listed for our pages are long and sentence-shaped, such as “google search central hiring seo no one guarantee ranking”. The page with the most Bing impressions was our Local Ranking Factors 2026 guide, with 93 impressions. DHM DATA The pattern could reflect assistants running searches, but Bing does not label queries by source, so we cannot confirm that. DHM INTERPRETATION

Google’s Knowledge Graph Search API returned no match for our business name on October 21, 2026, although a Google search for the name does show our Business Profile panel. DHM DATA We have not measured our own mention rate yet. The check above is where we start, and with this few visits we could not tell from analytics whether AI sends us customers.

What not to do

• Do not judge from one answer. Two runs of one question can name different businesses. One good answer is not evidence of consistent visibility.
• Do not chase a ranking position. SparkToro found that the order of a list repeats even less often than the list itself, so its authors track how often a brand appears.
• Do not buy a visibility score without asking how it was calculated. SparkToro advises asking any provider for public, reviewable methods and math. A score based on a few runs has the same wide range as your own.
• Do not write a page for every question. Google says that creating separate content for every possible variation of how people search, primarily to manipulate rankings or generative AI responses, violates its scaled content abuse policy. One clear page per service and town is enough.
• Do not promise yourself a mention. No one can guarantee that an assistant names a business, and we do not.

None of this changes how local SEO works. A complete Business Profile, real reviews and clear pages still come first. What the studies add is a reason to measure by rate and range, so a good week or a bad week is not mistaken for a trend.