AI Marketing for Window and Door Companies: How to Win Serious Homeowner Leads

01The category's physics: a three-week decision

Window and door marketing is often managed as if it were roofing with different photographs. It is not. Roofing has an emergency branch — the leak at 11pm — that produces same-day decisions and rewards being the first credible name. Windows and doors have almost no emergency branch at all. The purchase is triggered by a slow accumulation of annoyances: a draught, condensation between panes, a heating bill, a dated facade, a door that no longer seals.

The published homeowner research is consistent on this. This Old House's 2026 survey of 1,000 US homeowners who had replaced windows found 68% spent one to four weeks researching companies and 42% researched for at least three weeks; 84% obtained two or more quotes and 41% obtained three or more. Houzz's 2026 study puts windows as the second most common exterior project at 20% of respondents, behind roofs at 22%, with a median spend of $7,000.

That timeline has one commercial consequence that most window and door marketing ignores. You are not competing to be called first. You are competing to be on the shortlist that gets called at all. By the time the phone rings, the homeowner has already read something — an AI answer, a directory card list, a comparison page, a cost page. If your name was not in the answer, you are competing against a shortlist you were not on, and the only way in is price.

02Where window research now happens

Three surfaces matter in this category, and they behave differently from each other.

Google's classic local results. Still the largest source by a wide margin. One vendor analysis of the window and door category reports that AI Overviews appear on only about 2.5% of window and door queries, against over 15% in other industries — a single secondhand vendor figure, so treat it as an indication rather than a fact, but it is directionally plausible: AI Overviews are least common on commercial, transactional and local queries, and windows sit in all three.

Assistant research sessions. ChatGPT, Perplexity and Gemini get used for the questions a homeowner would otherwise ask a friend: vinyl or fibreglass, what should this cost, is it worth it, what rebates apply, who should I hire near me. These are precisely the questions where an answer naming two or three companies forms a shortlist. The click-through economics are documented: in a September 2026 study of 31,470 ChatGPT responses, home-services users clicked the first business named 12.1% of the time, the second 7.2%, the third 5.0%, and a further 8.4% searched the first brand's name instead of clicking.

Directories and review platforms. These remain the backbone of what assistants cite locally. Yext's analysis of 6.8 million citations found 48.73% of ChatGPT citations come from directory sources — a vendor figure reported secondhand, but consistent with everything else published. Which directory matters depends on country: for Canadian homeowners the stack is HomeStars-led, while American answers lean on Angi, Expertise.com and Yelp. An audit imported from the US will measure the wrong list.

03Position economics for a considered purchase

The position gradient is steeper than it looks, because a window buyer is doing this once. There is no repeat behaviour, no habit, no brand loyalty carried in from a previous purchase. The first name in the answer gets the benefit of the doubt by default, and the study data above says the first name is clicked roughly 2.4 times as often as the third.

There is a second-order effect specific to this category. Because buyers compare quotes, they read criteria, not just names. A homeowner who has read that ENERGY STAR certification in Canada means a U-factor of 1.22 W/m²K or lower will ask every company on the shortlist what their units rate. If one company's site answered that question in the research phase and the others did not, the sale is half-decided before the appointment. In a comparison category, the content that supplies the buying criteria wins the comparison.

04Five prompt classes that produce window leads

The prompt mix for windows differs from roofing in one important way: the research classes dominate. A homeowner asking about a leaking roof is looking for a phone number. A homeowner asking about window cost is building a mental spreadsheet she will use to judge your quote.

Classes 1 to 3 are where a window and door company can out-earn its market position, because they are answerable with facts you already own: your price book, your product specs, your rebate knowledge. Class 5 is the hardest and the least controllable — it is decided by third-party data about you. Class 4 is the one to publish anyway, because honest payback content is what an assistant pulls when it needs to talk someone out of a bad decision, and that is a citation worth having.

05The page inventory

Six pages carry almost all of the AI-search value in this category — a shorter list than the general window company marketing advice usually recommends, because most of that advice predates assistants reading the same pages. Everything else on a window company site — gallery, showroom, about, financing brochure — is support material.

Before any of that, check the basics. In an October 2026 scan of 3,198 US home-service websites, only 22% passed all five AI-readiness checks, 52% had no LocalBusiness-type structured data and 61% had no FAQ content on the homepage — and the study singled out remodeling businesses as the weakest trade, with 47% using LocalBusiness markup and 29% answering questions on the homepage. Window and door companies sit inside that group, and so do the adjacent categories — garage doors run into the same readiness problems, as our garage door company marketing work keeps finding. A rebate table on a site with no structured data still helps, but it is doing the work alone.

06The rebate wedge: why dates beat adjectives

Here is the opportunity almost nobody in Canadian windows has taken. The single most decisive, most frequently asked, most falsifiable question in the category is "what rebate can I get on windows" — and a large share of contractor websites answer it with programmes that closed a year or two ago.

The federal picture in Canada is closed on both counts. The Canada Greener Homes Grant stopped accepting new applicants in February 2024, with the final document deadline for existing files on 31 December 2025. The 0% Canada Greener Homes Loan closed to new applicants on 2 October 2025. Its successor, the Canada Greener Homes Affordability Program, is income-tested and delivered through named provincial partners — and as of late 2026 Ontario was not among them.

The live money is provincial, and it is smaller and more conditional than the old federal numbers most sites still quote.

Compiled November 2026 from provincial programme pages as reported by multiple contractor and consumer sources. This is the most volatile table in the article: treat amounts as indicative and confirm with the administrator before a homeowner relies on them.

Two technical details decide eligibility in Canada and almost no contractor site states them. First, ENERGY STAR certification in Canada is a single national standard since 2020: a U-factor of 1.22 W/m²K or lower, with a separate Most Efficient tier at 1.05 or lower. Ontario's rebate requires the certified tier, not the Most Efficient one — an error we see repeated in Toronto marketing copy. Second, rebates are paid per rough opening, not per pane: a bay window made of three sections in one opening counts once. Both facts are quotable, both are checkable, and both change the arithmetic a homeowner is doing in her spreadsheet.

07Energy specs as extractable facts

Window marketing copy is written in adjectives: advanced glazing, superior thermal performance, industry-leading efficiency. Every one of those phrases is unfalsifiable, and unfalsifiable claims are exactly what a retrieval system cannot use. The numbers, by contrast, are ideal citation material.

Six specifications do most of the work in this category. Put them in a table on the page, in text a machine can read:

Two rules when you publish these. Attribution per product: a U-factor belongs to a specific unit with a specific glazing package, not to your company. State the model and the package, or the number is marketing rather than fact. Honest trade-offs: a page that says a high-SHGC unit keeps a south-facing living room warmer in February and hotter in July is more useful — and more quotable — than one claiming blanket superiority.

08There is no WindowInstaller type. Here is what to use.

A technical wrinkle worth stating because it comes up in every window and door audit we run: schema.org has specific types for RoofingContractor, Plumber, Electrician and HVACBusiness under HomeAndConstructionBusiness — but no window or door installer subtype. Companies that try to force one invent invalid types, which is worse than using a general one.

The workable pattern, in order of value:

• HomeAndConstructionBusiness for the business, with name, telephone, address, areaServed as a city list, openingHoursSpecification, priceRange and sameAs pointing at every profile you own.
• Product on product pages, with additionalProperty entries for U-factor, ER and SHGC. This is how a numeric spec becomes readable rather than buried in a PDF data sheet.
• Service on service and product-plus-city pages, scoped with areaServed — window replacement, door replacement, patio doors, bay and bow conversions.
• FAQPage only where the question and answer genuinely appear on the page. Marking up questions that are not visible is a policy violation and, more practically, an inconsistency an assistant will resolve against you.
• WarrantyPromise and Offer if you can maintain them. Both rot fast if they are not generated from the same data as the page.
• Do not expect schema to lift citations on its own. The best-controlled public test — 1,885 pages that added JSON-LD against roughly 4,000 matched controls — measured −4.6% in AI Overviews, +2.4% in AI Mode and +2.2% on ChatGPT. Schema reduces ambiguity; it does not create authority.

09The economics of an in-home quote

Window and door has a cost that roofing mostly does not: the sales appointment. A proper whole-home consultation runs ninety minutes plus travel, with samples, financing options and a price presented at the table. At a weak close rate you are running a dozen of those to sell one job, and the estimator's loaded hourly cost is a marketing expense hiding in the operations budget.

That changes what "lead quality" means. In this category, a lead that will never book an appointment is not cheap at $45; it is expensive at any price, because the cost of discovering it is an hour of a salesperson. Which is why the ceiling should be set from the job backwards:

Run the formula with your own numbers and the strategy usually announces itself — it is the same ceiling arithmetic we apply in window and door lead generation before recommending any source. If your average job is a two-window repair at a low four-figure value, the ceiling per lead is small, appointment cost dominates, and you should be buying nothing but exclusive high-intent traffic and referrals. If your average job is a whole-home replacement, the ceiling is high enough to justify a research-content programme, because the homeowner's three-week research phase is worth more to you than the click that ends it.

10Lead sources compared, honestly

Every figure in this table comes from someone selling leads or marketing, which is exactly why the ranges are so wide. Use it for ordering your effort, not for forecasting.

Note the blank. Nobody has a defensible close rate for AI-sourced window leads, because the volume is too small and the attribution too broken to measure. Anyone quoting one is guessing. What we can say is that the channel's output in this category is disproportionately brand search and directory calls — which is why the instruments in the next section matter more than the referral report. For the broader channel mix in this vertical, see windows and doors marketing.

11Conversion mechanics for a considered purchase

A window buyer is not in a hurry, is comparing three quotes, and is afraid of being sold to. That produces a different set of conversion mechanics from an emergency category.

• Two calls to action, not one. "Book a free in-home estimate" for the ready buyer and "Get the cost guide" for the researcher in week one. In a three-week process, most first-time visitors are not ready for an appointment, and forcing one costs you the second visit.
• Publish the price range. The homeowner expects it, the assistant may be quoting it, and withholding it to protect the appointment is a strategy built for a market that no longer exists — 32% say price is the single most important factor.
• State the quote-accuracy commitment. 27% of homeowners paid more than quoted. A page that says what is included, what is excluded and what triggers a change order is a genuine differentiator, and it is quotable.
• Do not lead with financing. Only 15% financed in 2026, down from roughly 30% a year earlier; 46% used a credit card and 34% paid cash or cheque. Offer financing, do not build the campaign on it.
• Show the installation crew. Homeowners compare installers, not just products, and the survey data says installation quality matters as much as price. Named crews, certification and tenure beat stock photography.
• Answer the rebate question on the page where you ask for the appointment. "We prepare the documentation" is a reason to choose you that costs you nothing to say and is verifiable.
• No countdown timers on a $7,000 considered purchase. They read as pressure, and this buyer's main fear is being pressured into the wrong decision.

12Measurement: quotes, stages and windows of time

Three measurement problems are specific to this category, and each has a fix.

Problem one: the decision is longer than the attribution window. A homeowner researches for three weeks, collects quotes over two more, and signs in week six. Default analytics windows will credit the last click — often a branded search — and ignore the AI answer that started it. Fix: report on a 90-day lookback for this category, and keep a separate brand-search curve by city as the leading indicator.

Problem two: the quote is the real event. A lead that never becomes a quote is worthless in this category because it consumed an appointment. Fix: instrument the funnel as lead → appointment booked → appointment held → quote issued → contract signed, and report conversion at each step. The step most companies never measure is appointment held, which is where no-shows quietly destroy the economics of expensive lead sources.

Problem three: the channel hides. A homeowner who reads an AI answer, then searches your brand name, then converts from organic, will be credited to organic search. Fix: the intake source field. One question — "How did you first hear about us?" with "an AI assistant" as an option — plus call tracking with dynamic number insertion, and a note in the CRM of the answer given. It is self-reported attribution, it is unfashionable, and it is the only instrument that sees this channel at all.

# The five columns your weekly reporting needs. Nothing else.
week | AI-referrer sessions | branded search (city) | appointments held | contracts signed

If your dashboard cannot produce those five columns, the first marketing project is the dashboard, not the content. The measurement guide for high-intent leads sets out the full instrumentation in more depth; it applies to this category with one substitution — replace "jobs installed" with "contracts signed".

13The 90-day build

Before week four, it is worth listening to the advanced digital marketing strategies for window and door companies episode, which covers the appointment economics this build assumes. And one sequencing note that matters more in this category than in roofing: the rebate page goes early, not late. It is the fastest citation to earn and the fastest to lose to a competitor who updates before you do.

14Reviews and shortlist formation

In a comparison category, reviews do two jobs: they get you named, and they are read while the homeowner is building the shortlist. The thresholds are consistent across the studies we track.

BrightLocal's 2026 consumer research reports 68% of consumers will only use a business rated four stars or higher, 47% will not use one with fewer than 20 reviews, 74% weigh only reviews from the last three months, and 89% expect the owner to respond. The same four-star figure shows up in the Canadian home-services benchmark data for this category, which puts window and door among the most review-sensitive verticals in home improvement.

The tactical implication for a business with a 90-minute in-home sale: ask for the review at the moment the crew finishes, and ask for a description of the work — which rooms, how many units, how the crew behaved in the house. That language is what an assistant quotes and what a comparison shopper recognises as real. A five-star review saying "great job" adds a count and nothing else.

15Five things not to buy

• "Top of ChatGPT" guarantees. Nobody controls retrieval, the same question asked twice on the same day produces different sources about 59% of the time, and any vendor promising a fixed position is selling something that does not exist.
• Monthly AI-visibility dashboards with no CRM join. A citation tracker is a diagnostic. If the deliverable cannot be reconciled to appointments held and contracts signed, it is a slide, not a channel.
• Pay-per-lead volume packages sold to a 90-minute appointment model. The math rarely works, and the vendor keeps the margin whether the appointment is held or not.
• Rebate content written by a copywriter with no programme knowledge. A wrong amount is worse than no page, because the homeowner discovers it at the kitchen table and the whole quote loses credibility.
• Financing-led creative as the primary angle. 3% of homeowners name financing as the most important factor. Test it; do not build the year on it.

16Sources and confidence

Every figure traces to one of the following. Where the source is a company with a commercial interest, it says so.

Two warnings. The rebate table is the most perishable content in this article and the most consequential if wrong — programmes close and amounts change, and the correct behaviour is to verify with the administrator, not to trust a blog. And almost all of the conversion and citation data is US-derived; the Canadian specifics here are the rebate stack, the ENERGY STAR standard and the HomeStars-led directory layer.

17Questions we get asked

18Glossary

U-factor

Whole-unit heat loss rate. Lower is better. Canada's ENERGY STAR threshold is 1.22 W/m²K; Most Efficient is 1.05.

ER

Energy Rating, a Canadian metric combining heat loss, solar gain and air leakage. Programme thresholds commonly reference 34.

SHGC

Solar heat gain coefficient: the share of solar heat a window admits. Higher means more winter gain and more summer heat.

Rough opening

The framed hole a window or door occupies. Rebates are paid per rough opening, not per pane, so a three-section bay counts once.

NAFS / CSA A440

The North American and Canadian performance standards that rate a window assembly for wind, water and air.

Prompt class

A group of AI queries producing the same kind of answer and requiring the same page and conversion path.

Card list

The business cards an assistant renders beneath a local recommendation. The selection surface in this category.

Brand-search follow-up

Users who search the recommended brand instead of clicking. Measured at 8.4% for the first-named home-services business.

HomeAndConstructionBusiness

The schema.org type to use for a window and door company, since no installer-specific subtype exists.

Appointment held

The stage where a booked in-home consultation actually takes place. The step that quietly decides the economics of expensive lead sources.

Cost per sold job

Channel spend divided by signed contracts. The only denominator that matches a window company's P&L.

Held-out panel

A fixed set of prompts asked in the same order every week, producing a position log instead of a screenshot.