I took 216 real U.S. home service websites — HVAC, plumbing, roofing, solar, remodeling, restoration — and ran every one of them through the same seventeen checks an AI assistant effectively applies before it will name a business in an answer. 210 sites came back readable. Here is what they look like from the machine's side.
The short version: the failure is not technical. These sites are, on the whole, perfectly legible to a crawler. Almost none of them block AI bots. Most have structured data. What they don't have is anything specific enough to repeat.
HOW THIS WAS MEASURED
- SAMPLE
- 216 U.S. home service companies from my own prospecting list — a convenience sample, not a random one, skewed toward small and mid-size independents rather than national franchises.
- METHOD
- One request to each homepage and one to its robots.txt, from an identified crawler, in August 2026. Seventeen checks scored pass / partial / fail and weighted into a 0–100 figure. Same code that runs the public scanner.
- NOT MEASURED
- Whether any assistant actually names these companies. This measures the inputs a model needs, not the output. Off-site signals — reviews, directory listings, mentions — are outside what a page scan can see, and they matter at least as much.
- EXCLUDED
- Server response time and page weight are reported but left out of the conclusions: the batch ran from Europe, so latency figures reflect distance, not hosting. 6 sites could not be read at all (2 behind bot-protection firewalls) and are excluded from all percentages.
The distribution
Average 72.3 out of 100, median 75. That is higher than I expected, and the reason is mundane: modern site builders ship LocalBusiness schema, headings and a canonical tag whether or not the owner knows what those are. The floor has risen. The ceiling has not.
AI readiness score, 210 sites
LANDINGPEAK, AUG 2026
Only 18.1% land in the band where an assistant has everything it needs. The bulk sit in the middle: named when a question is easy, skipped when it is competitive.
Where it actually breaks
Split the seventeen checks into what they're really asking, and the shape of the problem appears immediately.
Average score by group
LANDINGPEAK, AUG 2026 — 210 SITES
Machines can read these sites. Machines can broadly trust these sites. What machines cannot do is quote them.
That gap — 82 for readability against 55.9 for quotability — is the whole finding. Every dollar the industry spends on technical SEO is being spent on the column that is already fine.
answer no customer question anywhere on the page, in any form
never state hours, emergency policy or how fast they respond
carry no dollar figure at all — not a range, not a starting price
Read the middle one again. These are emergency trades. Roughly 40% of the businesses whose entire value proposition is we come out tonight do not say so in text a machine can find. When a homeowner asks an assistant who can come out tonight, that sentence is the answer — and it isn't there.
What separates the top from the bottom
Take the 38 sites scoring 85 or above and the 53 scoring under 65, and compare them check by check. The ordering is not subtle.
Share passing each check
TOP 38 (85+) VS BOTTOM 53 (UNDER 65)
Every one of these is a decision about what to say, not a decision about how to build. The largest single gap in the whole dataset is real price figures: 86.8% of the top group against 3.8% of the bottom.
A 83-point gap on publishing prices. Not schema markup, not site speed, not a framework — whether the business is willing to put a number on the page. That is a commercial decision most contractors have already made in the opposite direction, on the theory that a price scares people off before the phone call. Against an assistant, that theory costs you the mention entirely, because "contact us for pricing" is not a sentence any model can repeat to a homeowner.
What I expected to find and didn't
I went in with three assumptions from the industry conversation. The data killed all three.
NO MEANINGFUL DIFFERENCE BETWEEN TOP AND BOTTOM
- AI crawlers allowed — 92.1% of the top group, 88.7% of the bottom. A +3-point difference, which is to say none.
- Title tag — 78.9% of the top group, 79.2% of the bottom. A +0-point difference, which is to say none.
- Page weight — 23.7% of the top group, 35.8% of the bottom. A -12-point difference, which is to say none.
The one about crawlers deserves its own line, because it is repeated constantly — including by people selling AI visibility services. Only 5 of 210 sites, 2.4%, disallowed any AI user agent in robots.txt. A further 2 sat behind bot-protection firewalls that turned away an identified crawler outright, which is a real problem for those 2 businesses. But the picture of an industry accidentally locking GPTBot out at scale is not what this sample shows. If someone opens with that as your main problem, ask them to show you your robots.txt first.
The page-weight result is the one that surprised me most. The top group's pages were heavier — a median of 2099 words against 540 in the bottom group. Lean pages score badly here for the obvious reason: there is less on them. Speed still matters enormously for whether a human sticks around, and I've written about what slow costs you. It is simply not the thing that decides whether a model can describe your business.
By trade
Modest spread, and I'd treat the ordering as directional rather than definitive — these are sample sizes in the tens.
Average score by trade
TRADES WITH 8+ SITES IN SAMPLE
| Trade | Sites | Average | Median |
|---|---|---|---|
| Roofing | 51 | 75.6 | 79 |
| Solar | 16 | 74.4 | 75.5 |
| Other trades | 21 | 72.9 | 73 |
| HVAC | 50 | 72.1 | 75 |
| Plumbing | 28 | 69.2 | 71 |
| Remodeling | 32 | 67.8 | 67.5 |
The interesting thing is how narrow the range is. There is no trade where this is being done well — only trades where slightly more of the builders happened to ship decent defaults.
What I'd do with this if it were my company
Ranked by the size of the gap this data actually measured, not by what's easiest to sell.
None of that is a rebuild. Five of the six are writing tasks that the owner is better placed to do than any agency, because they require knowing what a water heater swap actually costs in your county.
The one-sentence version
- Average 72.3/100 across 210 live contractor sites — the floor is higher than people think
- Readability 82, quotability 55.9 — the gap is the entire problem
- Publishing real prices separates the top group from the bottom by 83 points
- 2.4% block AI crawlers — that scare story is mostly not true
- Only 8.6% have a price, a response time and an answered question all on one page
If you want your own number rather than the average: the scanner is free and runs the identical seventeen checks against your homepage, or against you and a competitor side by side. The reasoning behind each check, with the outside studies it rests on, is in the 2026 GEO field guide.
If any of these numbers are useful to you, take them. The only thing I ask is a link back, so the next person can check the method for themselves.
PLAIN TEXT
Landingpeak (2026). AI Visibility Index: 210 U.S. home service websites scored for AI readiness. https://landingpeak.com/blog/ai-visibility-index-2026
HTML — WITH THE LINK ALREADY IN IT
<a href="https://landingpeak.com/blog/ai-visibility-index-2026">Landingpeak AI Visibility Index 2026</a> — 210 U.S. home service sites scored for AI readiness
Anonymised: trade and state kept, company names, domains and cities removed. These are real businesses and several of them are people I would like to work with — the point is the pattern, not naming anybody.