Why Does ChatGPT Mention Your Competitors but Not Your Roofing Company? The 87% Rule

The 60-second answer

The first number is the one worth remembering, because it moves the problem somewhere you can act. If a competitor's presence were decided by the quality of their website, the answer would be "build a better website". It is not: it is decided by whether the sources the engine consults happen to include them. That is a distribution problem, and distribution problems are solved by being present in more places, not by writing better pages about yourself.

You are not absent, you are being substituted

There is a difference between not appearing and being replaced, and the distinction matters because it changes what to look for. A roofing company that is genuinely absent does not appear in the answer and does not appear in its sources either — the engine never encountered it. A company that is being substituted appears in the sources and is not carried into the answer, or fails to appear in the sources for a question a competitor answers instead.

The shapes are worth separating because the industry conflates them, and because they have different costs. The absent shape is cheapest to fix and most common. The substituted shape is the most irritating, because the brand is doing the work and taking none of the credit — in one June 2026 study of AI citations, 62% were ghost citations where the domain was used and the brand never named, 25% were brand mentions without a citation, and only about 13% carried both.

The third shape is the most expensive and the least common, and it is where the position problem begins. Only 36 brands out of more than 1,200 tracked across four AI platforms held top-100 visibility in every month of a four-month study — and even within that group mention volume varied more than thirty-fold. For a local roofing contractor, the realistic ambition is not universal visibility; it is being the name that resolves for one question in one city.

The 87% rule: the decision precedes the answer

The mechanism that explains most of this question is documented in a single measurement. SearchIntel's second study collected 12,264 searches that ChatGPT, Claude and Gemini wrote for themselves while researching 177 category buying questions across UK and Irish markets. Those internal searches named a specific company on 52% of the questions in one market and 66% in the other. And when a company was named at that step, the final answer mentioned it 87% of the time.

Read the arithmetic. The engine does not hold a neutral shortlist and then choose fairly. Its research step already produces a reading list, and a reading list that names companies is a de facto shortlist. Once your competitor is on it, their inclusion is close to decided. You are not losing an argument in the answer; you are failing to appear in the material the answer is assembled from. The mechanics of how that material is fetched, split and selected are set out in the retrieval engineering brief.

The same study carries a second finding that rewards patience: between consecutive captures of identical questions, only about 47% of the names in Google AI Overviews appeared in both, 47% for AI Mode and 55% for Gemini. The shortlist is re-decided constantly. That is why a competitor's grip on a sentence is usually less permanent than it looks — and why a programme measured monthly will look like noise while a programme measured quarterly shows real movement.

Whose words are actually in the answer

Before diagnosing why you are not in the answer, it helps to know whose words constitute the answer. Here the published research disagrees in a way that is instructive rather than confusing, because the disagreement is entirely about definitions.

These figures are not contradictory so much as differently cut. Strictly owned pages account for roughly a quarter to two-fifths of the evidence base; everything else is third-party. Whether a Google Business Profile counts as "yours" or as "earned" changes the headline number and nothing else.

The reconciliation is the useful part. Your own website is a minority contributor, somewhere between a quarter and a sizable minority, and everything else — listings, directories, reviews, supplier pages, association entries, community threads — is a majority. So a contractor who has spent three years improving a website and nothing on the surrounding record has been optimising the smaller half of the problem, and the competitor being named is very likely winning in the larger half.

Five reasons a competitor gets the sentence

Almost every real case is at least two of these in combination, which is why single-cause explanations sold as fixes disappoint. The diagnostics in section eleven will tell you which ones apply; the fixes are not interchangeable.

One caution before the detail. A sixth cause exists and is not fixable: the competitor may hold a genuinely locked position in your market, with consensus across phrasings. That is a flanking problem rather than a gap problem, and section fourteen handles it separately. Diagnosing it early prevents a quarter being spent on a head-on assault that the evidence says will not work.

Reason one: the source gap

This is the most common answer to the question in the title, and the easiest to verify. The engine's sources for your category include the competitor because someone wrote about them. Not advertised them — wrote about them: a supplier's case study, a trade magazine round-up, a local news piece about a storm season, a community thread where residents named the roofer who fixed their leak.

Roofing has an unusually rich supply of these, which is good news. Four source classes do most of the work, and each produces the kind of prose that gets quoted rather than the kind that gets linked.

Cheers Market Baseline, 2,176 answers, Sep 2026; KKV Labs 60-city index, Sep 2026. Both vendor research. The Reddit and forum figure is the one most often quoted at conferences and the least often measured in a specific local market.

The diagnostic is blunt and takes an afternoon. Take five prompts where a competitor is named and you are not. For each, record every source the answer cites, and then classify each source: does it describe the competitor, and does it describe you? The result is a map of where you are missing. If the same three sources appear against four of five prompts and none of them mentions you, the mechanism is not ambiguous — it is a coverage hole, and no amount of on-page work closes it. Which of those sources are worth chasing, and in what order, is the subject of the roofing AI-SEO strategy set.

Reason two: your record is ambiguous

The second mechanism is the opposite of a coverage hole: you are everywhere, in slightly different forms, and the engine cannot tell whether those forms describe one business or several. The practitioner evidence for this is unusually direct — in Whitespark's 2026 Local Search Ranking Factors survey, 47 local search experts ranked the signals driving AI visibility and three of the top five were citation factors: name, address, phone and description matching across the web.

For a roofing company the ambiguity is usually manufactured by marketing, not by neglect. Call-tracking numbers placed on directory listings to measure source performance split one business into many. A trading name used on the website and a legal name on the licence registry create two entities. A service area that appears as a radius on one page, a list of cities on another and a Business Profile service area on a third gives three different answers to "where do you work".

This reason is under-diagnosed because it produces no drama. There is no competitor to blame and no tool that flags it; the site looks fine and the rankings are fine. It is also the cheapest of the five to fix — the sequence we run is documented in the AI SEO case study — typically three hours with a spreadsheet and a phone call to whoever owns the listings, and it is a precondition for two others — precision cannot compound if the same narrow claim appears in inconsistent forms, and freshness cannot help if the fresh evidence is attached to a different version of your business.

Reason three: they own a narrower claim

The third mechanism is a positioning problem wearing a technical costume. A recommendation is a match between a described need and a described business, and broad descriptions match nothing in particular. The competitor being named is usually not better; they are more specific.

Watch how this plays out in a real market. Ask an engine for a roofer and the generalists compete on the head term, where the directory wins because a directory is a better answer to a broad question than any single contractor is. Ask specifically about an insurance claim after a hailstorm, or a flat roof on a commercial strip, or a heritage home with a slate roof, and the answer changes — because now the question resolves to whoever describes themselves as that thing.

The discipline is to choose the narrow claim where your proof is already strongest — the most reviews mentioning it, the most documented jobs, the most third-party corroboration — and to state it identically on the site, in the Business Profile category, in the supplier locator and in the first line of every mention you earn. Precision only compounds when it repeats. A contractor who is first choice for insurance-claim roof repair in one city and fourth for everything else will out-earn one who is fourth for eight things, because the narrow question arrives with the decision already made — the commercial logic of which is in the high-intent leads breakdown.

Reason four: their proof is fresher

The fourth mechanism is time. Verification is not a one-off; the system is deciding whether a claim still holds, and evidence that moved last quarter is worth more than evidence that moved two years ago. Three clocks run at once, and the competitor is usually ahead on at least one of them.

Freshness also interacts with the instability described in section three. Because only about half the named businesses survive between consecutive captures, the source set is being re-read all the time — which means a competitor who has not published anything since 2024 is quietly exposed, and a contractor who establishes a publishing and review rhythm now will start appearing in gaps that open on their own. That is the most underrated argument for maintenance over campaigns: the shortlist turns over, and turnover is opportunity for whoever is current. Reviews do part of that work, and being named at all is the prior step covered in the recommendation piece.

Reason five: they are on the platform your buyer uses

The fifth mechanism is a channel problem, and it is invisible if you measure "AI visibility" as one number. The platforms do not agree with each other about who exists.

The practical consequence is that a contractor can be well known to one engine and unknown to another, and the reason is the source mix. Perplexity leans on BBB more than others — 59.6% of its roofing answers in one panel. Gemini leans on manufacturer programmes — 56.3%. ChatGPT cites about fifteen sources per response, which means breadth of presence matters more there than depth in one directory. A local market where buyers use several engines requires presence in each engine's preferred source classes, which is why cross-platform coverage is the most under-exploited advantage available: in one Florida study of 745 queries across five platforms, 97.2% of named roofers appeared on only one platform and none appeared on all five.

So the answer to "why does it mention them and not me" sometimes has nothing to do with your reputation at all: your buyer asked Perplexity, and everything you have built is legible to ChatGPT.

The teardown: reading an answer as a document

Diagnosis requires a fixed procedure, because the temptation is to read the answer emotionally. Run this on five prompts where a competitor is named and you are not, and record it in one table.

After five prompts, the table usually resolves to one or two mechanisms with almost no ambiguity. If the same supplier page appears in four answers and names only the competitor, that is a source gap. If your domain is cited twice and your name never appears, that is ghost citation and an attribution problem. If every answer names you third behind the same business, that is preference, and the position article is the right diagnosis. If the list changes entirely when the prompt is reworded, nobody holds the position and the fastest route is precision rather than displacement.

Two hygiene rules make the table trustworthy. Record the answer text verbatim, because paraphrasing loses the exact sentence you need to compete with. And run each prompt three times on separate days, because with only about half the names surviving between captures, a single run is a sample of one.

What to do about each one

The fixes are not interchangeable and the order matters, because two of the five are preconditions. Ambiguity recorded in your own listings will blunt everything done afterwards; staleness cannot be fixed by publishing while reviews are frozen.

A note on sequencing that saves quarters: fix the ambiguity first even when it looks trivial, because it is the only mechanism that is entirely within your control and it changes whether the rest is measurable. Then close the single largest source gap identified in the teardown, not the easiest one. A supplier page that names you is worth more than three directory profiles nobody cites.

Why a better website loses

The belief that the question implies — that the best roofing company should be named — does not survive the data. Being named is close to decoupled from conventional ranking quality, and the decoupling is measured rather than asserted. In a June 2026 cross-industry study of more than 1,300 recommendations, businesses in a Google top ten were recommended about 30% of the time; recommended businesses ranked in a top ten only 5% of the time; and recommendation rates for ranking first versus eighth were 31% versus 30%.

Two conclusions follow. First, a competitor with an inferior site can be named ahead of you, because the mechanism is not a site comparison. Second — and this is the part contractors find harder to accept — your own good site is not being under-rewarded; it is being used for something else. It is evidence for citations, which is a real outcome and sometimes sufficient, particularly in Google's AI Mode where one panel found the company's own site supplied 87.2% of home-services citations.

So the correct framing is not "our website is better, therefore we should be named". It is "our website is better, and it is competing in the citation layer, while the recommendation layer is being decided among third-party sources we have not entered". That is a fixable problem, and it is a different one from the problem most agencies sell.

Taking the sentence without a fight

Sometimes the competitor's position is genuinely locked — one cross-industry study found roughly 78% consensus for the leading local recommendation across differently worded prompts. Charging that with a better website wastes a year. Four routes work instead, and all of them are available to a roofing contractor.

The test for which situation you are in takes an afternoon: run three differently worded versions of the same question. A stable winner means a locked leader and a flanking strategy. A rotating winner means the position is contested, and direct competition — precision plus corroboration — is viable. Deciding this before committing a quarter is the single highest-value hour in the whole programme.

The twelve-week displacement plan

Two expectations to set honestly. The first movement is usually from absent to cited — your domain appearing in sources before your name appears in prose — so an early report may show citations rising with no change in mentions. That is progress, not failure; the second step is attribution, and it takes another quarter. And because only about half the names survive between captures, week twelve should be treated as a midpoint rather than a verdict.

How to tell it moved

Most AI-visibility reporting measures the wrong unit. For this question, four numbers matter and they should be kept apart rather than blended into a score.

Two statistical facts govern how these can be reported. At a 45% hit rate, a sixty-prompt panel yields a 95% interval of roughly 32% to 58%, so movements of a few prompts are noise; and because consecutive captures share only about half their names, a single week cannot demonstrate anything. Report a range across at least four weekly observations, with an untreated cohort of prompts beside the treated one, and state the prompt set in the report rather than in an appendix.

What does not close the gap

Sources and confidence

Two cautions. The strongest single finding in this article — the 87% rule — comes from a vendor study in UK and Irish markets, not North America, and it measures a research step rather than a disclosed ranking mechanism; treat it as a well-observed pattern rather than a specification. And the "how much of my narrative is mine" question has no agreed answer, because every study defines ownership differently; the defensible statement is that third-party sources account for a clear majority, which is enough to justify the work.

Questions we get asked

Glossary

Substitution

The pattern where a competitor is named for a question your business could answer, usually because they are named in the sources the engine reads.

Source gap

The absence of your business from the publications and platforms an engine trusts in your category, while competitors appear in them.

Ghost citation

A citation in which the domain is used as evidence without the brand being named in the answer.

Attribution

The step from having a passage used to having the business named in the sentence; the difference between cited and credited.

Share of answer

Of all brand names appearing in a tracked set of answers, the proportion that are yours. Distinct from share of voice, which counts presence across search surfaces.

Mention rate

The share of tracked prompts in which the business is named; averages are low and vary sharply by market.

Source class

A category of evidence — directory, supplier, editorial, review platform, forum, own site — used because engines rely on them in different proportions.

Capture overlap

The share of named businesses that appear in both of two consecutive observations of the same question; roughly half in published trackers.

Locked leader

A market where the same business is named first regardless of prompt phrasing, indicating that flanking is more productive than direct competition.

Flank

Winning a specific version of the question — emergency, insurance, material, segment or adjacent municipality — rather than the head term.

Precision claim

A single sentence naming trade, work type and places, narrow enough that the business is the obvious match rather than a plausible one.

Ambiguity audit

Comparing name, address, phone, category and service area across every external record and removing contradictions.

Evidence base

The full set of sources describing a business, of which its own website is a minority contributor.