Original measurement
For some trades the assistant reads the businesses’ own websites. For others it never opens them.
755 citations across 175 answers, 37 trades, three cities. Overall 53% of what the assistant cited was a business’s own website - and that average is the least useful number on this page. In twelve trades the share was 80% or higher. In twelve others it was 20% or lower, and in the weakest of them it was 0 of 22. Which side your trade falls on decides whether work on your own website can reach the answer at all.
0 of 22
citations that pointed at a business’s own website, in the trade where the assistant reads them least (assisted living communities)
Key findings
- Across 755 citations in 175 AI answers about local businesses, 53% pointed at a business’s own website and the rest at directories, review sites, professional bodies and publishers.
- In twelve of the 37 trades we measured, 20% or fewer of the assistant’s citations went to a business’s own website; in the weakest, 0 of 22.
- In twelve trades the opposite held: 80% or more of the citations went to the named businesses’ own sites.
- Between trades, the share of an assistant’s citations pointing at a business’s own website runs from nothing at all to every single one; the site-wide average of 53% describes no trade in particular.
How this was measured
- When
- Readings taken 31 August and 1 September 2026.
- Instrument
- Our own gate rig: a real Chrome instance driven over the DevTools protocol against a logged-in free ChatGPT account. Every answer is stored whole, with its citation pills, before anything is counted.
- What counts as a citation
- Every citation pill the answer rendered, read from the link it points at rather than from the text printed on it. Duplicates within one answer are counted once.
- Classification
- All 459 distinct cited domains were sorted by hand into five kinds: a single business’s own website, a directory or marketplace, a professional body or government list, a publisher or forum, and a search utility. The classification ships with the data, one row per domain, so any call can be checked or disputed.
- Threshold
- A trade appears in the table only once it has at least 8 citations behind it.
What we measured
We asked an assistant the kind of question a customer asks - best dentist in Austin, roofer in Austin, assisted living in Austin - and recorded not only which businesses it named but which pages it read to answer.
The question behind it is the one every buyer of this category should ask first: when the assistant builds that answer, does it open the businesses’ own websites, or does it only ever read somebody else’s page about them? If it never opens theirs, then improving theirs cannot change the answer, however well it is done.
The result, trade by trade
Across 755 citations in 175 readings, 53% pointed at a business's own website and the rest at a directory, a review site, a professional body or a publisher. That average hides the finding: the shares do not spread evenly across trades, they cluster at the two ends. In 12 trades we measured 80% or above; in 12 others, 20% or below. Whether that split belongs to the trade or to the city is a separate question, and one we have not settled.
| Trade | Own site | Own of all citations | Readings | Cities |
|---|---|---|---|---|
| Concierge & DPC doctors | 100% | 14 / 14 | 3 | 1 |
| CPA firms | 100% | 27 / 27 | 5 | 1 |
| Addiction treatment centers | 96% | 26 / 27 | 6 | 3 |
| GLP-1 & weight-loss clinics | 94% | 17 / 18 | 3 | 1 |
| IV therapy lounges | 94% | 15 / 16 | 3 | 1 |
| IT providers (MSPs) | 93% | 14 / 15 | 3 | 1 |
| TRT & hormone clinics | 92% | 12 / 13 | 2 | 1 |
| Pet boarding & grooming | 91% | 20 / 22 | 3 | 1 |
| HVAC companies | 87% | 13 / 15 | 3 | 1 |
| Ketamine clinics | 86% | 19 / 22 | 4 | 1 |
| Private-pay therapists | 83% | 10 / 12 | 2 | 1 |
| Cold plunge & sauna studios | 80% | 8 / 10 | 2 | 1 |
| Financial advisors (RIA) | 79% | 34 / 43 | 6 | 3 |
| Laser & skin clinics | 79% | 15 / 19 | 3 | 1 |
| Home care agencies | 78% | 29 / 37 | 5 | 1 |
| Acupuncture clinics | 77% | 20 / 26 | 5 | 3 |
| Orthopedic groups | 76% | 16 / 21 | 6 | 1 |
| Insurance agencies | 74% | 20 / 27 | 5 | 1 |
| Fertility clinics | 71% | 10 / 14 | 3 | 1 |
| personal trainers | 57% | 8 / 14 | 3 | 1 |
| Driving schools | 43% | 6 / 14 | 3 | 1 |
| Gyms & training clubs | 38% | 3 / 8 | 3 | 1 |
| Nutritionists & dietitians | 38% | 9 / 24 | 6 | 1 |
| Pilates studios | 35% | 6 / 17 | 3 | 1 |
| massage studios | 33% | 7 / 21 | 5 | 3 |
| Cosmetic dermatologists | 17% | 4 / 23 | 8 | 3 |
| Kitchen & bath remodelers | 14% | 3 / 22 | 6 | 3 |
| Roofing contractors | 13% | 3 / 23 | 5 | 3 |
| law firms | 12% | 2 / 16 | 3 | 1 |
| Dentists | 11% | 1 / 9 | 3 | 1 |
| Funeral homes | 10% | 2 / 20 | 6 | 3 |
| Marketing agencies | 10% | 1 / 10 | 4 | 1 |
| Med spas | 7% | 1 / 14 | 5 | 3 |
| Assisted living communities | 0% | 0 / 22 | 5 | 3 |
| Interior designers | 0% | 0 / 9 | 4 | 1 |
| Moving companies | 0% | 0 / 21 | 6 | 1 |
| Plastic surgeons | 0% | 0 / 19 | 6 | 1 |
Own site = citations pointing at a single business’s own website. Everything else - directories, marketplaces, review sites, professional bodies, licensing boards, city pages, publishers and forums - counts against it. Trades with fewer than 8 citations are not shown.
The instrument we threw away first
Our capture stores each citation twice: the pill’s visible label, and the link behind it. A convenience field in the raw file holds a domain list built from the labels, and the first version of this measurement used it.
It keeps between a quarter and a third of the citations. What it keeps is not random: a pill whose label is already a bare domain survives, and a pill labelled with the business’s name is dropped. Reading the links instead raised the evidence behind this page roughly fourfold.
We then assumed that field must under-report own sites, since those are the pills most likely to carry a business name. We measured it rather than assuming, and we were wrong: it runs a few points high, not low. The reason to abandon it is that it discards most of the data, not that it leans one way. This is the third classifier we have written and discarded on this dataset, and the reason all of them are named here is that a measurement you cannot check is not a measurement.
What we could not establish
Whether this belongs to the trade or to the city is not established. Our repeat readings outside Austin are one per trade. Across the 10 trades measured in more than one city, the rank correlation between the Austin share and the out-of-town share is 0.59; at ten trades a correlation has to reach about 0.65 before it means much, so this neither establishes the split as a property of the trade nor rules it out. The extremes hold in every city we read; the middle of the table does not. Readings that rendered a business list also cite own sites more often than readings that did not (59% against 40%), so part of a trade's score is whether a list renders for it at all.
What this means if you are buying
- If your trade sits at the top of that table, your own website is the surface the assistant actually reads, and work on it is the lever that moves the answer.
- If your trade sits at the bottom, no amount of work on your page puts you into that answer, because the assistant never opens it. What decides the answer there is which directory, review site or professional register it does read, and how you appear inside it. That is a different job from the one most of this category sells.
- Anyone offering to optimise your site for AI answers without first establishing which half your trade is in is guessing, and half the time the guess costs you the whole engagement.
Honest limits of this measurement
- Whether this belongs to the trade or to the city is not established. Our repeat readings outside Austin are one per trade. Across the 10 trades measured in more than one city, the rank correlation between the Austin share and the out-of-town share is 0.59; at ten trades a correlation has to reach about 0.65 before it means much, so this neither establishes the split as a property of the trade nor rules it out. The extremes hold in every city we read; the middle of the table does not. Readings that rendered a business list also cite own sites more often than readings that did not (59% against 40%), so part of a trade's score is whether a list renders for it at all.
- One assistant, one account type, and a free account cannot pin the model tier, so the tier is recorded per reading rather than chosen.
- A cited domain we have not classified is counted on neither side - not as an own site and not as a gatekeeper - and excluded from the fraction entirely, because assuming an unknown domain is a directory is exactly how a figure like this quietly becomes wrong. 0 citations are excluded on that basis here. If unclassified domains ever exceed one citation in twenty, we do not publish these figures at all until they are sorted.
- The classification of 459 domains is ours and is a judgement in a handful of cases. The eight we were least sure of are marked as such in the published file; counting every one of them the other way moves the overall share by two points and changes no trade’s position.
- A trade’s share is not a forecast for one business in it. It says what the assistant read across the answers we captured, not what it will read tomorrow about you.
Questions about this research
Does this mean my website does not matter?
It means it may not matter for this particular channel, in this particular trade. Your site still does every other job it does. What the measurement says is narrower and more useful: in some trades it is not the page the assistant reads when it answers, so changes to it cannot show up there.
Which half is my trade in?
The table above covers the trades we have enough citations for. If yours is not there yet, we do not know, and we would rather say so than guess. Running the reading on your own business and its city is the way to find out, and that is what we sell.
So should I just pay the directories instead?
Not automatically. Which directory the assistant reads varies by trade, and a paid placement in the wrong one buys nothing. The order that makes sense is to measure first, see which sources the answer actually cites for your trade in your city, and then decide where the effort is worth spending.
Does a business have to be in the map list to be cited?
No. We counted citations to businesses whether or not that same answer put them in its map list, and answers with no map list at all still cite businesses. Answers that did render a list cite own sites more often (59% against 40%), which is one reason we treat the middle of the table as unsettled.
More measurements from the same instrument
- An AI answer to a business owner cites 8.7 sources
- How many companies can an AI assistant name at all?
- All seven measurements →
We run this measurement for a living. If you want it run on your own category and your own city, the free scan is the same instrument on one question, and the report is the full set. Nothing here is behind a form.
Every category we measure has its own page, with the questions customers put to an assistant in that category and what a reading costs. See all industries we measure →
Published 31 August 2026. Every figure on this page is recomputed from raw observation files by a single script, so any of them can be traced back to the readings behind it. If you are named here and believe a reading is wrong, tell us: we will run the measurement again and publish what it returns, whichever way it goes.