Our methodology, in the open
How we measure (and improve) whether AI recommends your roofing company
When a homeowner used to need a roofer, they Googled one. Increasingly they ask an AI — ChatGPT, Google's AI Overviews, Copilot — "who's the best roofer near me?" and act on the answer. Getting into that answer is a different problem than ranking on Google, and most of what's sold to solve it is built on claims that don't survive scrutiny.
This page explains exactly how we measure AI visibility, what we do to improve it, and what we don't know and won't pretend to. We publish it in full because the field is full of confident claims that fall apart under questioning, and the fastest way to show we're different is to show our work.
Two layers. Plain for owners deciding if this is worth their time. Detail — sources, sample sizes, honest caveats — for anyone checking our reasoning. Read whichever you need.
The one thing to understand first: AI answers are not rankings.
Ask an AI the same question 100 times and you won't get the same answer 100 times. The list of companies it names changes from one ask to the next. This isn't a glitch — it's how these systems work. They generate a fresh answer each time rather than reading from a fixed list.
Why this matters for you: any tool or agency that tells you "your business ranks #3 in ChatGPT" is selling you a number that doesn't exist. There is no rank. ChatGPT doesn't have a position #3 for roofers in your city any more than a conversation does.
What we measure instead is frequency of inclusion — out of many asks, how often does your company show up? "You appeared in 7 of 10 asks this week, up from 4 last week" is a true statement. "You rank #3" is not. Every number we report is a frequency with the sample size attached, never a fabricated rank.
For the detail-minded
Industry analysis has found the probability of an AI returning the same brand list across two separate asks of the same question is below 1%. That's why we run each query in your panel multiple times per engine per cycle (currently five) and report the distribution. A single ask is anecdote; the distribution is data. Any vendor reporting a stable "AI rank" is either not running repeated trials or smoothing non-deterministic output into a number that implies a precision it doesn't have.
Your Google Business Profile is the spine.
When an AI needs to recommend a local roofer, it leans heavily on Google Business Profile data — your hours, categories, services, reviews, location. For Google's own AI Overviews, your profile is essentially the source of the answer. A neglected profile is close to invisible to AI regardless of anything else you do. It's also the most fixable thing on this list, which is why we start here.
For the detail-minded
A 2026 analysis of over 350,000 business locations found Google Business Profile to be the dominant data source AI systems draw on for local recommendations, and found that profiles not updated within roughly 30 days saw significant drops in how often they surfaced. Freshness is a continuous requirement, not a one-time setup.
Your reviews are a gate, not a dial.
This is the finding that surprises most roofers, and the one that determines whether we can even help you. For traditional Maps results, a 3.7-star business that's nearby still shows up. For AI recommendations, that's not how it works. AI systems appear to apply a rating floor — below a certain star average, a business is essentially excluded from recommendations entirely, no matter how good everything else is. Above the floor, more stars help, but the floor itself is pass/fail.
What this means practically: if your rating is below roughly 4.0, the honest answer is that no amount of optimization will get you reliably recommended until that comes up first. We'll tell you that directly rather than take your money and work around the edges. Sometimes the first real project is your reviews, not your AI visibility — and if so, we'll say so.
For the detail-minded
The same 2026 study found businesses recommended by ChatGPT averaged about 4.3 stars; for other engines, figures landed nearer 3.9–4.1. We treat ~4.0 as the practical working floor across engines. Two honest caveats we won't paper over: first, these are observed averages of recommended businesses and inferred confidence thresholds — they are not officially published cutoffs from OpenAI, Google, or anyone else, and we don't represent them as such. Second, these per-engine figures trace to a single major study, so we hold them as a strong directional input, not gospel. We also weigh review text — recency, whether reviews mention specific services, your response rate — because the star number isn't the whole picture above the floor.
What others say about you matters more than what you say.
AI systems weigh third-party corroboration — what shows up about you on Reddit, YouTube, industry directories, and review sites — heavily. Your own website matters, but the consistent picture of you across the web matters more. For roofing specifically, video is unusually powerful, and the bar is lower than you'd think. The work here is the slowest and least flashy. It's also the most durable, which is why it's where lasting AI visibility actually comes from.
For the detail-minded
Analyses of AI citations have found YouTube to be a standout source, with the large majority of cited videos being long-form, and a meaningful share having relatively small view counts — meaning substance and relevance, not virality, drive citation. The peer-reviewed work on what makes content more likely to be cited (a Princeton/KDD study) found that adding real statistics, quotations from named experts, and citations to authoritative sources measurably increased citation likelihood, while keyword-stuffing performed below baseline. We build content on those levers — substance, evidence, structure — not volume.
The honest part: what AI visibility is and isn't worth right now.
AI search is still small in raw traffic. Across the web today, AI tools drive a tiny fraction of visits compared to traditional search — the gap is large, on the order of a hundredfold. If someone tells you AI search is already a flood of traffic, they're overselling.
But it's growing fast, it's high-intent, and the recommended slots are scarce. A homeowner asking an AI "who's the most reliable roofer in town" is deep in buying intent, and the answer names only a handful of companies. For a high-ticket purchase like a roof, being one of those few names — early, before your competitors get there — is worth far more than the raw traffic numbers suggest. We think of this as positioning for where things are clearly heading, not a faucet you turn on for leads tomorrow.
Attribution is genuinely hard, and we won't fake it. When a homeowner calls, they rarely say "ChatGPT sent me." We can measure how often you're included in AI answers — that's our deliverable, and it's real. We cannot hand you a clean count of leads attributable to AI, because that data mostly doesn't exist yet. Anyone promising precise AI-lead attribution is promising something the current tools can't deliver.
This field changes in weeks, not years — so we monitor continuously.
The sources AI draws on shift constantly. In one documented stretch in 2025, one major platform's share of AI citations fell from around 60% to around 10% in about six weeks. Major model updates can reshuffle how recommendations work overnight. A technical setup that made you visible last quarter can quietly stop working.
This is why we don't do a one-time audit and walk away, and why we re-check the ground truth continuously rather than quarterly. It's also, frankly, why we built our own measurement system rather than renting one — so we can watch your visibility at the speed the field actually moves, and catch a shift the week it happens rather than the quarter after.
For the detail-minded
Independent analysis has found very high volatility in local AI recommendations — the set of recommended businesses churns continuously as models re-evaluate websites, reviews, and third-party sources. We re-run your measurement panel weekly per metro, re-test the technical fundamentals after major model releases, and track the mix of sources AI cites in your category so we can see drift early.
How we keep our own numbers honest.
A measurement system in a volatile field is most dangerous when it breaks silently and keeps reporting confident numbers that are wrong. We've designed specifically against that. We measure AI answers two ways: at volume through engine APIs, and through periodic checks of what a real person actually sees in the consumer apps. When those two disagree beyond a tolerance, that's a flag — it means our fast numbers may not reflect a homeowner's real experience, and we investigate before we report.
If our system can't stand behind a number, you don't get that number. We'd rather tell you a result is pending than ship you something we don't trust. Absence is safer than false confidence.
We also record what we saw and why we concluded what we concluded — not just the result — so that when we tell you something changed, we can show you the basis for it rather than ask you to take it on faith.
What we won't claim.
The absence of these claims is part of the point.
- We won't give you an "AI rank position." It isn't a real thing.
- We won't promise a specific number of leads from AI. The attribution data to back that up doesn't exist yet.
- We won't claim the review floors or any other figure here are official algorithmic rules. They're the best evidence available, held honestly as evidence.
- We won't pad your content with AI-generated volume. Thin, mass-produced content doesn't help and increasingly hurts.
- We won't tell you we can fix your AI visibility if your reviews are below the floor. We'll tell you to fix the reviews first.
Seeing where you stand.
If you want to know how often AI currently recommends your company — and where the specific gaps are — that's exactly what our initial audit shows: your inclusion frequency across the major engines for your real service-area queries, your profile and review standing against the practical floor, and the technical fundamentals that may be making you invisible. Many roofers are surprised by the result, including ones who rank at the top of Google Maps and assume that means they're covered. The two are not the same thing, and the gap is usually invisible until someone measures it.
Get your audit →We update this page as the field changes and as our own findings accumulate. Where we've cited research, we've described it plainly rather than dressing up estimates as certainties; where the evidence is thin or single-source, we've said so. If something here is out of date or you think we've got it wrong, we want to know.