AI Visibility for Roofing Companies: Storm, Replacement and the Facts AI Misstates

Roofing buyers ask two kinds of question. Storm prompts are urgent and local. Replacement prompts are slower and built around comparison. Both end in a shortlist of names, and the assistant decides who is on it.

Here are example prompts for a hypothetical roofer in Oklahoma City:

• "Who can tarp a storm-damaged roof in Oklahoma City today?"
• "Hail damage roof inspection near Edmond. Do they help with insurance claims?"
• "Best roofing companies for a full roof replacement in Oklahoma City"
• "Is [your company] licensed and insured in Oklahoma?"
• "What warranty does [your company] give on a shingle replacement?"
• "Does [your company] work in Norman and Moore?"
• "Metal roof or architectural shingles for a house in tornado country?"

Two things follow. First, these questions name facts the answer has to state, right or wrong. Second, the assistant rarely runs your prompt as typed. OpenAI says ChatGPT search rewrites a query into one or more targeted queries. So test the questions buyers actually ask, not keyword strings.

As of October 2026, five surfaces shape this buying journey. Each one pulls from its own index, so measure them one at a time.

• Google AI Overviews and AI Mode. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link, with no additional technical requirements. Your Google organic health is the gate.
• ChatGPT. Search answers depend on OAI-SearchBot reaching your site. Without search, answers come from training data with a fixed cutoff. OpenAI's model docs list a cutoff of August 31, 2025 for the Instant path, so a warranty change made this year may not exist in an ungrounded answer.
• Microsoft Copilot. It grounds on Bing. In Otterly's 2026 experiment, Copilot kept citing sites that Bing still indexed after Google had dropped them.
• Perplexity and Claude. Each uses its own search crawler (PerplexityBot and Claude-SearchBot).

Retrieval is discretionary on every platform. The same storm prompt can come back from live search one time and from memory the next. That is why trained knowledge and live search need different fixes.

No published study measures how often AI misstates roofers specifically. These four facts are where errors are most likely, because they are detailed, they change, and they often appear in different versions across the web.

• Warranties. Most roofers carry two: the manufacturer's material warranty and their own workmanship warranty. An assistant can merge them, attach the wrong term, or cite a manufacturer program you no longer hold.
• Licensing. License numbers, license types and states. An old directory listing or a lapsed number can surface as current.
• Insurance-claim help. "Helps with claims" can be read as "handles your claim for you". Vague wording leaves the model to fill the gap.
• Service areas. One storm job in another county can turn into "serves the whole state".

Some errors are outdated facts from training data. Others are invented. The fix differs, so diagnose which kind you have before you edit anything.

The one positive experimental signal for on-page content is narrow: a specific fact, stated in a sentence an engine can quote, on a page it can fetch. Write each fact that way, in HTML that loads without JavaScript.

Hypothetical wording for the Oklahoma City roofer:

• "Our workmanship warranty covers installation defects for [X] years. Shingle materials are covered separately by the manufacturer's warranty."
• "We are licensed in Oklahoma (license [number]) and carry general liability and workers' compensation insurance."
• "We inspect storm damage, document it with photos, and meet your adjuster on site. You file the claim with your insurer."
• "We serve Oklahoma City, Edmond, Norman, Moore and Yukon."

Use the same wording on the warranty page, the about page and the footer. Contradictions across your own pages hand the model two versions to pick from. This does not guarantee a citation. It means a retrieved page carries the right sentence. When the old version persists in trained answers, follow the steps for correcting outdated facts past a knowledge cutoff.

When engines cite other sites instead of yours, being named correctly on those sites is associated with appearing in the answer. Nobody has measured the size of that effect, so treat this as cleanup, not a growth lever.

For roofers, check these:

• Google Business Profile and Bing Places. Service areas, hours, emergency service and categories.
• Your state contractor license lookup. This is the record an assistant may repeat.
• Manufacturer contractor locators. Whether your certification is listed, current and at the right tier.
• BBB, Angi, Yelp, HomeAdvisor and Nextdoor. Business name, phone, license and service area.
• Local news and storm-recovery lists. Whether old coverage names a service you no longer offer.

Fix each listing at the source, then wait for a re-crawl. You cannot force it. The same directory audit applies to HVAC companies and plumbers, who share most of these listings.

The tempting move after a hailstorm is fifty "Roof Repair in [Town]" pages. Google's spam policy defines scaled content abuse as many pages generated to manipulate rankings without helping users, and its first example is using generative AI to produce them.

Otterly tested volume in 2026 with two new domains of roughly 1,000 AI-generated posts each. Google deindexed one after about 14 days and the other after about 90. Otterly reports no manual action in Search Console on either. After removal, Google's AI surfaces almost never cited them. It was one pair of low-stakes sites, so it is not proof of what happens to a roofer. It does show the removal can be silent.

AI Overviews need an indexed page. Lose the index and you lose those citations too.

Do this instead:

• One service-area page that lists every town you serve in plain text.
• A town page only where you have real local substance: completed jobs, photos, a crew based there.
• No pages for towns you will not drive to.

Access levers have the strongest evidence of anything in AI search. These are things to check on your own site, not assumptions.

1. robots.txt per bot. GPTBot controls training only. OAI-SearchBot controls ChatGPT search. OpenAI says each setting is independent. A blanket AI block can remove you from ChatGPT search without anyone noticing.

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

2. Firewall and CDN. A firewall can return 403 to a crawler that robots.txt allows. From September 15, 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and free-tier customers, while Search stays allowed.
3. JavaScript. Vercel found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. Service-area maps, warranty tabs and financing widgets loaded by script may be invisible to them. Googlebot renders; most AI crawlers don't.
4. Snippet controls. A stray noindex, nosnippet, max-snippet:0 or data-nosnippet keeps content out of AI answers.

The free AI Crawler Checker from 99Visibility requests a page as each AI crawler and shows which ones your firewall turns away.

Week 1: access

Run the four checks above. Then check index status in Google Search Console and Bing Webmaster Tools separately. A healthy Google index tells you nothing about Bing.

Week 2: baseline

Run your storm and replacement prompts and save each answer word for word, with the date. The free checker on this site shows whether assistants name you for three buyer questions. You can also check for free whether ChatGPT mentions your brand. Mark every wrong warranty, license, claim or service-area statement.

Week 3: your site

Write the four facts as plain sentences. Make the warranty page, about page, footer and Google Business Profile say the same thing.

Week 4: directories and cleanup

Correct the third-party listings. Merge or remove thin city pages. Re-run the same prompts and note which answers cited sources.

Expect search-grounded answers to change before trained ones. Trained answers update only with a new model, and nobody can schedule that.

Does allowing GPTBot help ChatGPT recommend my roofing company? No. OpenAI documents GPTBot as a training crawler only. ChatGPT search uses OAI-SearchBot, and the two settings are independent. You can block GPTBot and still appear in ChatGPT search, as long as OAI-SearchBot is allowed and your firewall lets it through. Check both robots.txt and your CDN before assuming you are reachable.

Should I add schema markup to my roofing site for AI answers? It is not a lever for AI answers. Google says structured data isn't required for generative AI search. Ahrefs' matched test of 1,885 pages found AI Overview citations fell 4.6% after schema was added. Existing markup does no harm, so you don't need to remove it. Spend the time on plain-text facts and access checks instead.

How long until ChatGPT reflects my new warranty terms? It depends on the answer type. Search-grounded answers can reflect the change once OAI-SearchBot re-crawls the page. Answers from trained knowledge wait for a new model with a later cutoff, and no one outside OpenAI controls that timing. State the new terms clearly on a fetchable page so grounded answers have the right sentence to quote.

Do I need to hire an AI search service for a roofing business? Not to start. The access checks, the four fact statements and the directory cleanup are work a marketing lead can do in a month. A service makes sense if you lack the time or need ongoing monitoring across engines. Read [whether a small business needs an AI search optimization service](/questions/small-business-ai-search-service) before you sign anything that promises rankings.