When AI cited a feature, it named the business 96 times out of 100.
A link to your feature only helps if the answer around it names you. In this study it almost always did. In the latest round, 100 answers linked to a business’s own feature page, and 96 of those answers named that business. The four misses all trace back to the wording on the feature page, which makes them fixable.
Feature citations by business, latest round (Sep 15)
| Featured business | City magazine cites | Local weekly cites | Named when cited | Answers asked |
|---|---|---|---|---|
| Fee-only wealth manager | 23 | 7 | 30 of 30 | 132 |
| Cannabis dispensary B | 16 | 7 | 21 of 23 | 132 |
| Email-delivery SaaS | 13 | 0 | 13 of 13 | 172 |
| Cannabis dispensary A | 10 | 5 | 15 of 15 | 120 |
| Scheduling SaaS | 7 | 5 | 12 of 12 | 172 |
| Fractional CFO firm | 4 | 0 | 2 of 4 | 156 |
| SEO agency | 3 | 0 | 3 of 3 | 156 |
| All seven | 76 | 24 | 96 of 100 | 1,040 |
Source: tracking round of Sep 15, 2026, four AI tools. “Cites” = answers linking to the business’s own feature page. “Named when cited” = those answers that also mention the business. Gemini citations are counted where its redirect links could be traced to the page.
City-magazine citations rose about a quarter on the AI tools we could measure every round.
The headline count went from 40 city-magazine citations in the first round to 76 in the latest. Most of that jump comes from better measurement. Gemini hides its sources behind short-lived redirect links, and we could only trace them to the feature page from the second round on. On ChatGPT, Claude and Perplexity, which we could read in every round, the count went from 40 to 51.
City-magazine feature citations by round, all seven businesses
| Round | Perplexity | Claude | ChatGPT | Three tools | Gemini | Total |
|---|---|---|---|---|---|---|
| Aug 12 | 25 | 9 | 6 | 40 | not verifiable | 40 |
| Aug 26 | 30 | 17 | 9 | 56 | 17 (2 businesses) | 73 |
| Sep 15 | 32 | 13 | 6 | 51 | 25 (all 7) | 76 |
Source: three tracking rounds, 1,040 answers each. Round two is labeled by the date its capture finished. Gemini is only credited in rounds where its redirects still worked, so the Total column can’t be compared across rounds. The three-tool column can.
The local weekly was cited less. Its buyer’s-guide pages were cited 15, 13 and 24 times across the three rounds, and only four of the seven businesses have ever had one cited.
The biggest single gain belongs to the wealth manager. Its city-magazine feature went from 9 citations to 17 to 23, and its local-weekly page picked up its first 7. Every one of those 30 latest-round answers named the firm.
In this study, features showed up mostly on questions that already named the business.
Customers ask AI about local businesses in two ways. A brand lookup names you: “is this wealth manager any good?” A discovery question doesn’t: “best fee-only financial advisor in [city].” Discovery questions are where new customers come from. In this study, features showed up overwhelmingly on brand lookups, and discovery is where they had the most room to grow.
Which questions got a feature cited, latest round
| Question type | City magazine | Local weekly | Share |
|---|---|---|---|
| Brand lookup (names the business) | 61 | 22 | 83% |
| Local discovery (no business named) | 12 | 0 | 12% |
| Other discovery | 3 | 2 | 5% |
How often the businesses appear on local discovery questions
| Round | Answers naming the business | Rate |
|---|---|---|
| Aug 12 | 129 of 336 | 38% |
| Aug 26 | 145 of 336 | 43% |
| Sep 15 | 113 of 336 | 34% |
Source: the questions in each business’s set that name the city but not the business, pooled across all seven, four AI tools. We chose this as the study’s measure of new-customer discovery before tracking began. It has not risen, and the latest round is the lowest of the three.
One business shows what local discovery can look like. The SEO agency appears in 26 of 40 local discovery answers. On national buying questions without a city, it appeared in 1 of 52 answers at baseline. AI treats it as a strong local answer and barely knows it nationally.
Counting the publication’s whole site would have credited the wrong business.
The two dispensaries compete in the same market, and both have features. Across the three rounds, 18 answers to dispensary B’s questions cited dispensary A’s feature. A count of every link to the publication’s site would have scored those as wins for dispensary B. If anyone reports AI citations of your feature to you, ask whether they counted your page or the whole site.
Rule 2: never credit what you can’t see. When an AI tool hides the page behind a redirect we can’t trace, the citation is reported as unverified and left out of the count.
Rule 3: say zero out loud. Three of the seven businesses have never had their local-weekly page cited. Their report says so, in words, instead of leaving a blank.
A local newspaper that blocks AI crawlers got zero direct citations. Its awards still showed up second-hand.
The local daily newspaper in this study blocks every crawler behind these four AI tools in its robots.txt file (the file that tells crawlers what they may read), including the live-search bots from OpenAI, Anthropic and Perplexity. None of the 3,120 answers could quote it directly. Its influence arrived second-hand, through business websites and listicles that repeat “voted best by the daily paper.” Before you pay for a feature, ask the publication whether its site lets AI crawlers in.
Four things to ask for so a feature helps with new customers
2. Make sure the page is clearly about you. Features that list competitors or bury the business name lose the “named when cited” rate. Every one of the four misses in this study came from a page like that.
3. Ask about “best of” roundups. “Best of” category pages are the format AI tools pull into discovery answers. A publication roundup that links to your feature turns one citation into several.
4. Judge the feature on discovery questions. Total citations flatter. The number that tells you whether a feature brings in new customers is how often AI names you when someone asks for your service in your city without naming you.
What this study can and can’t tell you
No before-and-after. The features went live in June and early August, before our first round on August 12. There is no clean “before” picture, so this study describes what features do in AI answers. It does not measure how much visibility they created.
The same questions every round. Each business had the same 30–43 buying questions every round, locked before tracking began, asked on the same four AI tool versions. Nothing was changed mid-study.
Answers read by software, checked by hand. A language model read each answer against a fixed set of rules, and we reviewed its corrections by hand after each round. Software readers tend to over-credit the business being tracked, so we report whether the business appeared and which pages were cited, the two counts that leave the least room for judgment.
Gemini is partial. Gemini citations count only where its redirect links still worked, which was the latest round for most businesses. Round-over-round comparisons use the other three AI tools.
Does AI name your business?
The Snapshot asks ChatGPT and Google’s AI the questions your customers ask and shows you who gets recommended.