We check 17 things on a local business website. One decides where AI ranks you.
We scanned 125 small local businesses, then asked AI assistants who they recommend in 6 towns, and compared the two. Most of what gets graded is a door you walk through once. This is the part that keeps mattering afterwards.
- Of the 17 things we check, 1 both separates recommended businesses from ignored ones and orders the recommended ones: Where You Operate.
- 41% of our own score sits on checks that only open a door: real and necessary, and unable to move you once you are through.
- 5 of our checks showed no measurable effect on who AI names. They are named on this page rather than left out of it.
Open your website and read the front page as a stranger: does it say your business name, the towns you cover and the exact jobs you take, in plain text?
- 1
- 2
- 3
- 4
- 5
- Northside Plumbing
- Valley Rooter Co
- Corona Drain Works
- Ridgeline Plumbing
- Oakfield Pipe
Every AI readiness tool, ours included, hands you a list of things to fix. None of them tell you which items on that list are known to matter, because that takes a second measurement: going and asking the AI who it actually recommends, in real towns, and checking that against the grades.
We did that. 138 business-and-market pairs across 12 markets, auto repair, electrical, landscaping, nail salon, painting, plumbing and roofing in 6 towns, including the parts that argue against our own scoring.
The short answer
We grade 17 things. Then we ask AI assistants who to call in 6 towns, thousands of times, naming no business in the question. One of the 17 shows up in who gets recommended, and it is the same kind of thing: a plain fact about your business, written on the page as text a machine can read.
Our earlier studies say why. AI reads businesses' own sites rather than the directories listing them (where it looks), and names the ones whose site mentions the exact job asked about (what your website says you do). Your site is what it has to go on.
The one that showed up
- Where You Operate — The towns you serve, written out where a reader would look for them.
If you only do one thing, start with Where You Operate. Open your website and check it is there, on the page, as words rather than a picture or a promise to load later.
Getting named at all and ranking well are two different problems
Does AI mention you at all? That is a door, and most of what we check is a door. 99% of the businesses we measured already score full marks on SSL / HTTPS, so it cannot explain why one of them outranks another. It matters. It cannot rank you.
Once you are through, where do you land? Different question, different answers, and the one most people actually want. 41% of our own score sits on checks that only open doors.
A check you have already passed cannot improve your position, no matter how important it was. Fix the doors once, then spend your time on the things below.
All 17, and the job each one does
Every check, with what we measured it doing. Three of ours did not separate anybody and two ran the wrong way, and those rows are here too.
- Each mark: One dot per check, placed by both of its measurements at once.
- Left to right: Points of score. A check at +20 means businesses AI recommends average 20 points higher on it than businesses AI never mentions.
- Up and down: How closely the check tracks position among the businesses AI already names. Zero means not at all.
- What to look for: Far right and low is a door: it tells the two groups apart and then stops mattering, and that is where most of our weight sits. Top right is a check doing both — but check the dot's own note first, because a check almost everyone passes can sit high on a number computed from very few businesses.
Show all 17 rows, with both numbers
| Check | Job it does | Gets you named | Decides your place | Weight |
|---|---|---|---|---|
| Where You Operate | Both | +15.8 | 0.26 | 20 |
| Who You Are | Both, but not yet certain | +17.3 | too close to call (0.16) | 20 |
| Content Organization | Both, but not yet certain | +20.5 | too close to call (0.14) | 25 |
| URL Structure | Both, but not yet certain | +2.7 | too close to call (0.12) | 10 |
| Entity Consistency | Both, but not yet certain | +8.5 | too close to call (0.12) | 15 |
| Crawl Health | Both, but not yet certain | +11.0 | too close to call (0.12) | 40 |
| Structured Data | Both, but not yet certain | +13.6 | too close to call (0.10) | 20 |
| Google Business Profile | Gets you named | +25.4 | too close to call (-0.03) | 50 |
| Contact Accessibility | Gets you named | +12.0 | too close to call (-0.01) | 25 |
| Accessibility | Gets you named | +8.7 | too close to call (-0.00) | 15 |
| SSL / HTTPS | Gets you named | +6.7 | too close to call (0.11) | 25 |
| What You Do | Gets you named | +6.3 | too close to call (0.02) | 15 |
| Page Speed | Neither, in this study | +0.9 | too close to call (-0.04) | 5 |
| Agent Interface | Neither, in this study | 0.0 | n/a | 5 |
| Link Health | Almost everyone already passes it | -0.9 | too close to call (0.10) | 5 |
| Legal & Compliance | Runs the wrong way | -4.3 | too close to call (-0.02) | 10 |
| llms.txt | Runs the wrong way | -3.2 | too close to call (0.05) | 10 |
Read the second number if AI already recommends you and you want to move up. Read the first number if AI does not mention you at all.
The one that does both jobs
This one separates recommended businesses from ignored ones and orders the recommended ones, by more than the sample size alone would produce.
Six more lean the same way without clearing that bar: Who You Are, Content Organization, URL Structure, Entity Consistency, Crawl Health and Structured Data. Worth doing. Not proven.
What to actually change
It is just words on a page. It does not need a developer, a rebuild or a subscription.
The checks that measured nothing, and why they are still on your report
Page Speed, Agent Interface and Link Health did not separate anybody in this study. Legal & Compliance and llms.txt ran the other way: the businesses AI names score slightly WORSE on them.
They stay, and not out of sentiment. A check earns its place either by predicting something or by finding a real problem on your site, whether or not AI cares.
Legal & Compliance is the clearest case: near the bottom of that table, and the check that got a customer to publish a privacy policy and terms the next day. No coefficient was going to tell them to.
llms.txt is the other kind. It measures adoption of a standard almost nobody has adopted, so there is nothing yet to separate. We are not moving its weight; being early is the position we want when that changes.
A check with no measured effect on AI is not a check with no value. It is a check we cannot yet show you a number for.
How we measured this
Every business here was scanned the same way your own free scan runs: we fetch the page as a machine would, with no browser, and grade the 17 checks. Separately, we ask AI assistants who to call for a trade in a town. We repeat each question so one lucky answer cannot change the result, and we never name a business in the question. Joining the two gives 138 business-and-market pairs across 12 markets.
"Gets you named" compares businesses AI recommends against businesses in the same towns it never mentions. "Decides your place" is measured only among businesses AI already names, so it cannot be inflated by the gap between the two groups.
Scope
- 12 markets in 6 towns, all of them in Southern California, weighted toward the trades a homeowner calls. A national or a different-trade corpus will move these numbers.
- We tested all 17 checks against the same outcome at once. Run that many tests and roughly one will look convincing by luck. So here are both numbers: Where You Operate came in at 0.004, and a stricter bar that allows for 17 simultaneous tests would ask for 0.003. Read it as the strongest thing on this page rather than as settled, which is also why the near-misses are listed rather than buried.
- This is a measurement of what goes together, not of what causes what. No business was changed and re-measured.
- The numbers will move as the corpus grows. When they move enough to change the advice, we publish a new study rather than editing this one.
See where your own site stands
The free scan runs every check on this page against your website and tells you which group each one landed in for you.
2ee53d5, data fingerprint 414dd8ffcc42. These are the numbers as of that date, on that corpus. The web is being rebuilt around how AI reads it, so they will move. When they move enough to change the advice, we publish a new study rather than quietly editing this one. See every study we have published.