One directory turned up in three cities on two continents while its whole class held 0.7%

Two facts from our own readings sit awkwardly together, and the awkwardness is the useful part. In the ten city scan taken on 15 and 16 October 2026, with six buyer questions per city on both engines, goodfirms.co was among the most cited sources in New York at 3 citations of 12 readings, in Mumbai at 3 of 12 and in Delhi at 3 of 12. In the wide run of 16 October 2026, across 1,260 readings and 20,143 citations, review directories as a class took 0.7%.

One directory, three cities, two continents, and a class share under one point. Neither figure is wrong. They are answers to two different questions, and holding both is what stops a plan being written on half the evidence.

The two readings, and what each one is a reading of

	The city scan	The wide run
Dates	15 and 16 October 2026	16 October 2026
Shape	Ten named cities, six buyer questions each, both engines	210 questions per market, three markets, both engines
Size	120 readings, all usable	1,260 readings, all usable
Question type	Who does this work in this named place	Everything from problem stage to procurement depth
What it says about directories	One listing site among the top sources in three of the ten cities	The whole class at 0.7% of citations

Read the question type row and the apparent contradiction resolves. A directory holds a page per city per category. When the question names a place and a category, that page is a close match to the question. When the bank asks 210 questions per market about problems, prices, roles, comparisons and timing, the great majority of them have no directory page to match, so the class share collapses.

What that means for a listing you are paying for

• The shape of question decides the value. If your buyers' questions name a place and a service, a current listing is in the running. If they ask about a problem, a price or a comparison, it is not.
• The directory did not lead anywhere it appeared. At 3 citations of 12 readings in each of those three cities it was level with local sites rather than ahead of them, and in one of the Indian cities it shared the top with two domains covering that region only.
• Keep the profile accurate for its own sake. A listing with your current coverage and contact details is one fewer place for a stale fact about you to live, and human buyers read these pages on purpose.
• Budget it as a reach asset. The class share is 0.7% of citations on a run of 1,260 readings, and an earlier bank of questions put it at 0.8%, so the two independent measurements agree.

The city a directory does not win

In the same scan, the most cited source in one American city was a national software company's agency listing subdomain at 4 citations of 12, and in three other cities the top of the list was held by domains built around that one place. So a global listing site reaching the top three in three cities is a real result and a partial one. It appears where it appears and local pages appear everywhere.

The honest instruction from both readings together is to find out which shape your own buyers' questions take before deciding what a listing is worth to you. That is a measurement rather than an opinion, and it takes an afternoon on thirty questions.

Where these numbers stop

Twelve readings per city is a small reading and it describes how an answer about that city was assembled rather than any industry in it. Ten cities have numbers and no other city does: there is no figure here for a place we did not read. Both readings used two apps, ChatGPT and Gemini, and the three further surfaces our product covers were asked nothing in either. Every count is a domain in a source list and not a business named in an answer's text, which matters doubly here because a directory's own pitch is usually about the second thing.

We sell the measurement and the publishing work that follows, so the figure that reflects on us goes in as well: across all ten cities our own domain was cited zero times. The Google Search Console side of our record, with both windows and the export date, is our proof page.

How to test this for your own category in an afternoon

The whole disagreement above is reproducible on your own questions, and the result is worth more to you than either of our numbers.

• Write ten questions that name a place and your service, and ten that name a problem, a price or a comparison with no place in them.
• Ask all twenty on two named apps, with no company name anywhere in the wording.
• Record every cited domain and mark which of the two groups the question came from.
• Count how many directory pages appear in each group. That difference is the number that decides what a listing is worth to you.

Forty readings is an afternoon, and it answers a question that no published class share can answer for your business, because the answer depends entirely on which of the two groups your buyers actually type.

Questions about directories in AI citations

Which cities had the directory among their top sources?

New York, Mumbai and Delhi, at 3 citations of 12 readings in each, from the scan taken on 15 and 16 October 2026.

How can that sit beside a 0.7% class share?

The two readings put different kinds of question. Naming a place and a service is the one shape a directory page is built to match, and the wide bank of 210 questions a market is mostly made of other shapes.

Did the directory lead any of those three cities?

No. In each one it was level at 3 citations with other sources, including small domains covering that place or region only.

So is a listing worth buying?

Work out which shape your buyers' questions take first. Where they name a place and a service, a current listing is in the field. Where they ask about problems, prices or comparisons, 0.7% is the size of the contribution.

Which engines were read for both?

ChatGPT and Gemini for the city scan and for the wide run. The earlier of our two category runs, with its full method and source mix, is published as 597 readings across ChatGPT and Gemini in three markets.

Two numbers, two questions, one decision

The temptation with a pair like this is to pick the figure that supports what you already planned. Both belong in the same paragraph: a listing site reached the top three in three cities on two continents, and its whole class carried 0.7% of citations in a run of 1,260 readings. Which of those applies to you is decided by the questions your buyers actually type, and that is the thing to go and measure.