Generative engine optimization · Original data
The first source ChatGPT cites carries more weight, and directories get that slot 41.5% of the time
When ChatGPT replies to a local search, it doesn't name just one source. It names several, sometimes a dozen for the same question, and it always puts them in a sequence. We'd never paid attention to that sequence before. Every other time we've sliced this data, we just counted which websites appeared rather than where each one sat in the list. This time we looked at the order, and the source that shows up first ended up mattering more than we thought.
The data is our old standby: 1,423 citations that ChatGPT and Perplexity attached to 200 local-service questions across ten types of business, dentists, gyms, plumbers, hair salons, med spas, optometrists, auto shops, chiropractors, personal-injury lawyers, and vets, in five cities from Tulsa to New York, pulled on August 26 and 27. Every answer lists its sources in the exact order the engine gives them, spot one through spot twelve. This round, we didn't group by domain. We grouped by that number.
The first position plays by its own rules
Out of all 1,423 citations, 31.8% point to a directory or review site instead of an actual business. That's the split we've used before. Look only at position one, the very first source in each reply, and directories jump to 41.5%: 83 of the 200 questions we ran led off with a directory. The difference keeps showing up further down the line, too. Directories average position 3.94 across the whole set. Business pages average 4.62. Directories aren't just more common in general. They show up earlier, and the earliest spot is where they're most overrepresented.
How big that gap is depends on the category. For dentists, a directory takes the top slot 60% of the time, that's 12 of the 20 questions we ran. Auto repair shops and gyms come in at 55%. Personal-injury lawyers are at the opposite end: only 20% of their first slots go to a directory, the lowest of the ten, which means four out of five times a law firm's own page leads the answer. Plumbers and vets sit in the middle, in the low thirties. Run a med spa or a hair salon and you're right around the dataset average, a directory leads about 40% of your category's answers.
Why being first might carry extra weight
We went looking to see if that lead position actually influences a reader more, or if we were just imagining meaning in how the engine formats a list. A paper published September 15 by researchers Wang Claire, Agam Goyal, Frederick Choi, Koustuv Saha and Eshwar Chandrasekharan ran a controlled test with 278 people, showing them AI-written summaries and real customer reviews for the same products in different orders. Whichever source people saw first anchored their final opinion of the product, and only a very strong signal from the reviews was enough to pull that opinion away from the anchor. People also underestimated how much the AI summary had shaped their judgment. They said they trusted the reviews more, then rated the product as though the first thing they'd read mattered most.
That study looked at product pages on a shopping site rather than local businesses answering a search, and we're not saying the mechanism is identical. But the shape of the finding matches something we'd suspect even without it: whatever a reader sees named first in an AI answer does more work in their head than the fourth or fifth name on that same list. If that holds for ChatGPT's local-business answers too, then a directory sitting in the first slot 41.5% of the time isn't just a neutral fact about list sorting. It means the source most likely to shape the reader's first impression, in four out of every ten of these questions, isn't a business at all.
What this means and what it doesn't
We can't run the anchoring experiment on ChatGPT's own output, so the link between our position data and that study's finding is a solid guess rather than a direct measurement. The five cities in our set lean toward bigger metros, and a category built on 20 questions can swing on a handful of answers. Dentists at 60% first-slot directories means 12 out of 20 questions rather than 1,200, and running the same queries a different week could shift the numbers a few points either way.
What stands regardless of that caveat: directories aren't just cited more often than businesses, they're cited first more often than their overall share would predict, in nine of the ten categories we checked. If your business keeps landing third or fourth on the rare answer where it gets named at all, position might matter as much as presence. That could be the next thing worth chasing.
You can check where your own listings show up in about thirty seconds.
Source: Wang, Goyal, Choi, Saha, Chandrasekharan, "First Impressions: How Placement Shapes the Influence of AI Summaries" (arXiv, September 15, 2026), a randomized experiment with 278 participants comparing AI-generated and user-written product summaries in different presentation orders. Our own figures come from askchat.studio's GEO probe database: 1,423 citations across 10 local-service categories and 5 cities, collected August 26-27, 2026; position recorded as the order each engine listed its sources, unweighted by how the citation was used in the answer text.