Solar Installer Marketing: Local Pages and Case Studies Written With AI

A solar installer can use AI to draft a town page or an install case study from a short fact sheet, then check every claim against the company's own records before publishing. The AI supplies structure and wording. The proof, the system figures, the photos and the customer's consent have to come from real jobs.

Most AI-written town pages fail for one reason: the only thing that changes from page to page is the town name. Google's spam policies call that pattern doorway abuse or scaled content abuse, and fifteen near-identical pages tell a search engine nothing about your company. The fix is a fact sheet for each town and each finished job, filled in from your records before any AI is involved.

Follow me on Instagram@sagnikteaches

What Google's spam policies say about town pages

Two definitions in Google's spam policies matter here. Doorway abuse covers pages "created to rank for specific, similar search queries", and one of Google's examples is pages "targeted at specific regions or cities that funnel users to one page". Scaled content abuse covers many pages generated mainly to manipulate rankings, and its examples include "using generative AI tools or other similar tools to generate many pages without adding value for users".

Connect on LinkedInSagnik Bhattacharya

Google's separate guidance on generative AI content is more relaxed than people expect. It says AI can help "to add structure to original content" and asks publishers to fact-check and review everything it produces for accuracy before it goes live. So the line is not "AI or no AI". The line is whether each page carries something a reader could not get from the page next to it, and whether a person checked it.

Subscribe on YouTube@sagnikteaches

For a solar installer, "something a reader could not get elsewhere" is already in your files: the roofs you have surveyed, the system sizes you have fitted, the problems you meet on one estate and not another, how long your own diary takes from enquiry to survey. A page built from those facts is a different page in every town without any trick. A page built from the prompt "write a solar page for [town]" is not.

Decide which towns earn a page

Not every town needs its own page. A rule that works: a town gets a page when you can fill at least six of the ten rows in the fact sheet below with something true of that town and not of the one next door. Towns that fall short go on a single "areas we cover" page, which is honest and far easier to keep current.

Take a five-person installer with two crews that has worked in three towns. The figures in this tutorial for that installer are illustrative, so swap in your own.

Keep the service areas on your Google Business Profile in step with the pages you publish. The profile has its own limits on how many areas you can list, so read its help pages before planning more town pages than the profile can support.

The fact sheet that stops the AI inventing things

An AI model fills any gap with something plausible. The fact sheet closes the gaps first. It is a ten-row table you complete from your own records, and the prompt later tells the AI to use nothing else. Build one per town and keep it in the same shared folder as the page it produced, so the next person can see where every sentence came from.

Here is Town A completed. Every cell is an illustration of the level of detail to aim for, not a benchmark.

A five-block skeleton with a word budget

A town page does not need 1,500 words. It needs enough real facts to answer what a homeowner in that town would ask before booking a survey. Five blocks cover it, and only the last one can share wording between towns.

That totals 450 to 700 words. A page that stays at 450 words because the facts ran out is better than one padded to 1,200 with generic paragraphs about the benefits of solar, which are exactly the paragraphs that repeat across towns. The tutorial on fence company marketing with local pages, photos and posts applies the same idea to another trade.

Prompt one: draft the page from the sheet

Paste the completed table under the instructions. The rules in the prompt do most of the work: a rule to mark gaps instead of guessing, a ban on savings and environmental claims, and the block order. Run it in a fresh chat each time so earlier pages cannot leak into the next one.

You are drafting a service page for a solar installation company.
Use ONLY the fact sheet below.

Rules:
- Every number, date, product name and place name must come from the
  fact sheet. If something is missing, write [MISSING: what is needed]
  instead of guessing.
- No savings, payback, "free energy", environmental-benefit or
  guarantee claims. Any generation figure is "modelled", and its
  assumptions go in the same or the next sentence.
- British English, plain words, no hype. 450 to 650 words.
- Blocks in this order: direct answer (50 to 60 words); what we have
  fitted here; what usually decides the design on roofs here;
  questions from customers here (use the two questions in row 9 and
  answer from rows 3 to 7 only); how the survey and booking work.
- Do not invent customer quotes, reviews or names.

FACT SHEET:
[paste the completed table]

Below is a typical slip, shown as an illustration of what comes back even with those rules. The opening of the draft for Town A reads:

We have fitted 14 solar systems in [Town A] over the last 20 months,
from 3.6 kWp to 6.8 kWp. Homeowners here enjoy plenty of sunshine and
can look forward to big savings on their bills. Our experienced local
team has served the area for years and is fully accredited.

The first sentence is right because it is lifted from rows 1 and 2. The next two are the AI filling gaps with confident filler.

Prompts two and three: claim check and sameness check

Do both checks in a new chat, so the AI is not defending its own wording. The claim check turns "does this read well?" into a list you can tick. The sameness check catches the repetition that Google's scaled content definition is aimed at.

Below are a draft web page and the fact sheet it was written from.
List every factual claim in the draft (numbers, dates, places, product
names, accreditations, timings, comparisons, promises). For each, give
the sentence and the fact-sheet row that supports it, or write
NO SUPPORT. Then list any sentence that promises savings, payback,
guarantees, environmental benefit or a result. Do not rewrite anything.

DRAFT:
[paste]

FACT SHEET:
[paste]

On the Town A draft above, the check returns something like this (illustrative):

The sameness check needs two finished pages. Paste both and ask for a comparison.

Compare Page A and Page B. List every sentence in B that is identical
to a sentence in A, or differs only by a place name, a number or a few
words. Count them and give the count as a share of B's sentences.
Do not rewrite anything.

PAGE A:
[paste]

PAGE B:
[paste]

Work the sum yourself so the AI's count is not taken on trust. Suppose the Town B page has 31 sentences and the check flags 9 that match Town A apart from the town name. Nine out of 31 is 29 per cent. A workable house rule is to rewrite anything above 20 per cent, excluding the booking block, from Town B's own rows. If Town B has no facts of its own to rewrite from, that is the signal to fold it into the areas page.

Case studies: capture, consent, then draft

A case study is one job told properly, which makes it stronger than a town page and harder to fake. The order matters. Capture the facts on handover day, get written consent before drafting anything, then write from the job's fact sheet and the customer's approved words. Drafting first and asking later is how photos end up published that the customer never agreed to.

What to capture on handover day

• A photo set from your install checklist. The solar installation quality checklist already has you photographing the finished job, so one set serves both purposes. For publishing, add the full array from a distance and keep house numbers, vehicle plates, faces and documents out of frame.
• Three or four questions answered in the customer's own words. What made them decide, what worried them beforehand, what surprised them during the installation, what they would tell a neighbour. Ask permission before recording; written notes are fine. The tutorial on writing case studies from customer interviews with AI goes deeper on interview technique.
• The job facts. System size in kWp, panel count, roof type and faces, survey and installation dates, crew size and days on site.
• The modelled figure from the quote. Copy the annual generation and the assumptions beside it, exactly as printed on the signed proposal.
• Optional, 12 months later: the customer's own generation reading from their monitoring app, if they offer it.

The one-page consent form

Keep the form short enough that people read it. Each use gets its own tick box, so a customer who is happy with a roof photo but not a quote can say so.

Consent and privacy rules differ between markets, so have a qualified adviser look at the wording once, and keep each signed copy with the job record. The consent reference goes into row 10 of the fact sheet, which means the AI never sees a photo you may not use.

Prompt four: draft from the job sheet and approved words

Here is a filled job sheet, again illustrative: a semi-detached house with a tiled roof, east and west faces, a 4.2 kWp system made of ten 420 W panels (six on the east face, four on the upper west face), installed over two days by a crew of three. The survey found a chimney that shades the lower part of the west face in the afternoon, so the layout skips those positions. The existing consumer unit had no spare way, so it was replaced during the same visit. The quote modelled 3,650 kWh a year for an east and west layout, with the pitch and shading allowance stated. The customer approved the sentence "I wanted to stop guessing what the roof could do, and the survey answered it."

Draft a 450 to 600 word installation case study for our website
from the job sheet below. Use ONLY the job sheet and the approved
customer quote, word for word.

Structure: a plain headline describing the job (no benefit claims);
the situation in the customer's approved words; what the survey
found; the design decision and why; the installation (days, crew);
the modelled generation figure with its assumptions; a facts box
(system size, panel count, roof, dates).

Rules: no savings, payback, environmental or guarantee claims. Any
generation figure is labelled "modelled at quote stage". If the job
sheet lacks something, write [MISSING]. British English.

JOB SHEET:
[paste]

A typical first draft stumbles in two places, so scan for them before anything else.

• It writes "The family now enjoys lower bills and a greener home." Neither half is in the sheet. Delete it, because "lower bills" is a savings claim and "greener" is the sort of generic environmental wording that regulators in some markets now challenge. Replace it with the modelled figure and its assumptions.
• It describes the chimney as "a minor obstacle". The surveyor found it shades four panel positions, which is why the layout has six panels east and four west. Put the real consequence back, because that detail is what makes the case study believable to the next homeowner with a chimney.

Answering customer questions without promising savings

The questions on a town page are where AI-drafted copy most often drifts into promises. Answer from the rows you can prove and write each answer in a pattern that carries its own conditions.

The same discipline applies to the proposal itself, which is covered in the tutorial on adding AI-written cover letters and FAQs to solar proposals. If your modelled figures come out of design software, the tutorial on what solar design software does and does not do explains which parts of the output to trust and which to double-check.

Link, publish and measure

Each town page should link to the case study from that town if one exists, to the survey booking page, and to nothing that looks like a list of forty neighbouring towns. A long footer of town names gives the reader nothing, and it makes the town pages look like a set of search-targeted pages rather than part of a site people browse. Case studies link back to their town page and onward to the closest service page.

Put the real facts in the title and the opening paragraph: "Solar panel installation in [Town A]: 14 systems fitted since [month]" is more useful and more accurate than a keyword string. Add a "facts last updated" date at the foot of each page and diary a quarterly refresh of the fact sheet, because counts and median timings drift.

To judge whether a page works, open Google Search Console, filter the performance report to that page's address, and look at impressions, clicks and the queries that triggered it. Add a "Which page were you on?" field or a hidden source field to your enquiry form so enquiries can be counted per page, and a "How did you hear about us?" dropdown for those who skip it. Give a new page several weeks before judging. If a page still has almost no impressions after that, fold it into the areas page rather than adding more words. The tutorial on solar lead-qualifying chatbots picks up from there, once enquiries arrive.

Four ways these pages go wrong in practice

1. The repeated paragraph. A paragraph on "why go solar" appears on every town page because it was in the prompt template. The sameness check reveals it as a block of matching sentences. Delete it, or write one proper guide and link to it.
2. The invented testimonial. The AI adds a quote such as "Brilliant service, would recommend" because case studies usually have one. Nobody said it. Remove every quote that is not on a signed approval, and use the claim check to find them.
3. The location leak. Phone photos can carry the exact place they were taken in the file's hidden data. Publish a copy with that data removed rather than the original file, and check how your website platform treats uploads instead of assuming it strips it.
4. The stale figure. A page still quotes a modelled range from proposals made before the design software's assumptions changed. Date-stamp the sheet and refresh it each quarter.

A realistic time budget for the first page

These timings are estimates for a first page. They fall as your job records get tidier, because the fact sheet is the slow part.

A case study takes about the same once the handover-day capture is routine, with the consent form adding a few minutes on the day. Two pages a week is a sensible pace for a small team, and it keeps every page checked by someone who knows the jobs.

Questions installers ask before publishing

How many town pages should a small installer publish?

As many as you can fill with real local facts, and no more. Use the six-row rule: if you cannot complete at least six rows of the fact sheet with something true of that town, list it on one areas page. Five strong pages beat twenty pages that differ only by the town name.

Can AI write the customer's quote in a case study?

No. Use only words the customer said or wrote, tidied for spelling, and send them the final wording for approval before it goes live. An AI-invented quote is a fake testimonial, and it undermines every real quote on the page. Ask the AI to format approved words, never to create them.

Should a case study show what the customer saved?

Avoid it. Savings depend on tariffs, usage and the year, so they do not transfer to the next customer. Show the system size, the roof, the problem you solved, and the generation figure modelled at quote stage with its assumptions. Add recorded generation only if the customer shares it and you label it as one home in one year.

Do I need written consent to show a photo of someone's roof?

Treat a photo of a person's home as personal information and get written consent for each use, as the form in this tutorial does. Rules differ between markets, so ask a qualified adviser to review the wording once, and keep every signed copy with the job record.

Can I reuse the same FAQ block on every town page?

Reuse the process wording at most, such as how a survey works. Questions and answers that come from jobs in that town, such as shade, roof types or timings, should differ page to page. A fully shared FAQ is one of the quickest ways to trip the sameness check.

Further reads

• Can AI Write a Solar Sales Script? Discovery Questions and Objections — Reuse the same fact sheet to prepare survey-call answers.
• Solar Battery Quote Template: Options Explained With AI, No Backup Promises — Add battery options to pages without backup promises.
• AI Marketing for Contractors: Reviews, Case Studies and Local Posts — A wider case-study and review routine for any trade.
• Home Inspector Website: Local Pages and Sample Report Copy With AI — See local pages built from a different trade's records.
• AI for Solar Installation: Five Jobs a Small Installer Can Automate — Marketing is one of five jobs worth handing to AI.
• Solar Lead Generation: Cost per Lead, Cost per Survey and AI Follow-Up — What each enquiry costs and how fast to follow it up.
• Sign Company Marketing: Case Studies and B2B Emails With AI — A job sheet, case-study prompt, consent email, local page test and three outreach emails a sign company can use this month.
• Solar Installer Software: CRM, Design and Tracking, and Where AI Helps — Buy order, vendor prices and three priced stacks for a five-person solar installer, with what AI does in design, CRM and tracking, and what it does not.
• Insulation Contractor Marketing: Autumn Campaigns and Local Pages With AI — A fact-sheet method for autumn campaigns and local pages that AI drafts without savings figures, payback talk or paraphrased grant wording.
• Paving Marketing: Turn Job Photos Into Local Content With AI — A six-shot routine, captioning prompts and permission wording that turn each paving job into posts, a project page and a profile update.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Search Central spam policies (doorway abuse, scaled content abuse) and Google's guidance on using generative AI content, both read in November 2026.