Solar Permit Application: Can AI Draft It? What Installers Must Verify

Yes, AI can draft a solar permit application narrative, cover letter and completeness checklist from the design facts you supply. It must not produce or approve the plan set, calculations or code references. The qualified designer verifies those, and the installer checks the finished package against the local authority's own published requirements before filing.

The line that matters for an installer is between words and numbers. A residential permit package is mostly engineering (drawings, string sizing, structural data) wrapped in admin text (a narrative, a cover letter, a checklist, replies to comments), and AI belongs only in the wrapping. What a small installer needs is a repeatable way to turn a signed design into a clean package with fewer returns. Permit rules differ by area and change, so treat this as general information and confirm every requirement against the permit office's current documents.

Follow me on Instagram@sagnikteaches

Where AI helps in a permit package, and where it must not

A residential permit package has two kinds of content. Engineering content (drawings, calculations, equipment data, structural statements) is created or approved by the qualified designer and belongs to whoever signs it. Admin content (the narrative, the cover letter, the checklist, the replies to comments, the inspection list) is language organised around facts. AI is useful for the second and risky for the first.

Connect on LinkedInSagnik Bhattacharya

The rule that follows: AI may restate a fact from the design, never create one. If the narrative says "11.4 kW DC", that figure came from the design file, and the designer confirmed it. If the model produced it, the narrative is wrong however confident it reads.

Subscribe on YouTube@sagnikteaches

Build a fact sheet before you ask for any text

Hallucinated details come from missing facts. A one-page fact sheet, filled from the final design, gives the AI everything it is allowed to say. A coordinator can build it from the design export in about ten minutes:

• Job code (not the customer's name) and system type (roof-mounted, ground-mounted, with or without battery).
• Array: module count, module model, rated power per module, total DC size, roof planes and tilt/azimuth per plane.
• Inverter(s): model, count, AC output rating, whether string inverters, microinverters or optimisers.
• Battery: model, usable capacity, location, if any.
• Mounting: roof type, rail and attachment system, whether the roof is replaced or reinforced.
• Electrical: point of connection, main panel details, whether a service upgrade is part of the job, rapid-shutdown arrangement, where disconnects sit.
• Documents attached: list of sheets and data sheets in the package by number.

Keep the sheet as plain text in the same project folder as the plan set, and paste it into every prompt for that job. If your design software exports a summary, as covered in solar installer software compared, paste that instead and let the sheet carry only what the export lacks.

Drafting the project narrative

Here is a prompt that keeps the model to supplied facts. The figures are illustrative.

You are helping a solar installer's permit coordinator.
Write a project description of 120-180 words for a residential permit application.
Use ONLY the facts below. If a fact needed for a normal narrative is missing,
write [MISSING: item] instead of guessing. Do not cite any code, standard or rule.
Do not use marketing language.

FACTS
Job code: R-2291. Roof-mounted PV system.
Modules: 28 x 410 W = 11.48 kW DC, in 3 roof planes.
Inverters: 1 string inverter, 10 kW AC output.
Battery: none.
Mounting: asphalt shingle roof, rail-based attachment, no roof replacement.
Connection: load-side connection at the main service panel; no panel upgrade.
Shutdown: module-level rapid shutdown devices installed.
Attached: sheets E-1 to E-4, S-1, data sheets D-1 to D-5.

Sample output, and what to fix

An illustrative reply:

This application covers a 11.48 kW DC roof-mounted solar photovoltaic system consisting of 28 modules rated 410 W each, arranged on three roof planes of an existing asphalt shingle roof. The system uses one string inverter with a 10 kW AC output and module-level rapid shutdown devices. Modules are attached with a rail-based mounting system. The system connects on the load side of the existing main service panel; no panel upgrade is included. No energy storage is part of this application. Supporting documents are attached as sheets E-1 to E-4 and S-1 and data sheets D-1 to D-5.

Two checks. "A 11.48" should read "An 11.48", a small slip but one a reviewer notices. More importantly, 28 times 410 W is 11,480 W, so the sum agrees, but the designer should still confirm the design file really has 28 modules and not a late change to 27. Whenever a number appears, verify it against the file, not against the prompt, since the prompt may be stale.

The completeness checklist: paste the office's list, never memory

An incomplete package is one of the most avoidable causes of delay for a small installer, and a checklist built from the office's own requirements prevents most of them. Do not ask AI what documents a permit needs. Paste the office's published checklist (from its website or application form), your package's file list and the fact sheet, then ask for a gap analysis.

Below are (A) the permit office's required-documents list copied from its website,
(B) the files in our package, and (C) the fact sheet.
Produce a table: Requirement | Matching file | Status (Present / Missing / Unclear) | Note.
Use only A, B and C. If a requirement depends on a fact in C (for example, battery
documents only if a battery is present), say whether it applies and why.
Then list anything in B that A does not ask for.

The last line catches extra pages that slow a reviewer, such as marketing brochures stapled to data sheets. The "depends on a fact" rule handles conditional requirements, such as structural documents only when the roof is modified or battery documents only with storage. A filled-in row might read: "Requirement: Roof attachment details | File: S-1.pdf | Status: Present | Note: S-1 shows rail layout; confirm it includes fastener spacing, since the requirement text mentions it." That note sends the coordinator to the right sheet rather than giving false comfort.

Cover letter and submission email

The cover letter is the easiest win. A reviewer wants to see the job address (entered by you, not by the AI), the permit type, the scope in one sentence, the contact for questions and an index of attached documents. Ask the model for a template with placeholders and keep it in your project system. The index can be generated from the checklist output, so the letter, the checklist and the actual files agree. If you submit online through a portal, ask for a version of the scope sentence within the portal's character limit and check what that limit is on the page itself.

A before-and-after shows the point. Before: "Please find attached our permit application for a solar installation at the property, thanks." After: "Permit application for a roof-mounted PV system, 11.48 kW DC, no battery, no service upgrade. The package contains 10 documents listed below. For questions, contact [name] on [phone]. Please note the roof attachment detail is on sheet S-1." The second version helps the reviewer find what they are likely to check first, and it saves a back-and-forth.

Plan-check comments: wording from AI, answers from the designer

When the office returns comments, the delay is usually the clarity of your reply, not the difficulty of the fix. Build a response table with three columns: comment (verbatim), response, and where it is addressed (sheet and revision). Paste each comment together with the designer's technical answer and ask for a concise, specific reply. The designer's answer is the input; the AI polishes the language and keeps a consistent tone. Say the comment reads "Provide labelling for the rapid shutdown initiation device." The designer's note says "Label added at the main panel and the inverter, sheet E-3 rev B." The reply becomes: "A rapid shutdown initiation device label has been added at the main service panel and at the inverter. See sheet E-3, revision B." Do not let the model add an assurance the designer did not give, such as "the system complies with all applicable requirements".

Track the exchange the same way you track the permit itself. Tracking permits and inspection dates with AI covers a simple log with submission date, comment date, reply date and approval date, so you can see which office is slowest and plan crew bookings accordingly.

An eight-permit month at a five-person installer

Say a five-person installer files about eight residential permits a month. A coordinator spent roughly 90 minutes per package assembling text, a checklist and a cover email, plus an average of one correction round per three packages. With AI drafting from a fact sheet, the same coordinator spends about 35 minutes on each package: ten for the fact sheet, ten to run the drafts, fifteen for the designer's read of the narrative and checks. Over eight packages that saves around seven hours a month of coordinator time, in this illustrative case, and the real gain is fewer rejections for missing documents. The saving evaporates if the fact sheet is built from memory instead of from the final design, which is why the first step is non-negotiable.

Cost of the tool is small. A paid chat plan with project features is typically enough, and business plans do not use business content for training by default, but placeholders for names and addresses remain the safer practice. If you already assess design software on cost per proposal, see solar AI design software compared by cost per proposal before adding another tool for the same job.

Inspection readiness: the list the crew lead uses

A permit approved on paper still fails if the inspector arrives and the labels, access or paperwork are not ready. This is another admin document AI drafts well, because it is a list built from your design facts and the inspector's published expectations. Paste the fact sheet and any inspection guidance from the office, and ask for a one-page checklist for the crew lead to tick on the morning of the visit.

Using only the fact sheet and the inspection notes below, write a one-page
pre-inspection checklist for the crew lead. Group items: Paperwork on site,
Labels and signage, Access and safety, Equipment matches plans, Photos to take.
Tick-box format. If an item depends on whether a battery or a service upgrade is
present, include it only when the fact sheet says so. Do not add any requirement
that is not in the notes.

A filled-in fragment, illustrative: "Paperwork on site: approved plan set (current revision), equipment data sheets D-1 to D-5, permit card displayed. Labels: inverter and main panel labels as shown on sheet E-3. Equipment matches plans: 28 modules counted, inverter model matches sheet E-1. Photos: array before covering wiring, label close-ups, inverter nameplate." The photo list matters twice over: it is evidence if the inspection fails, and it feeds the job's closing file. Where the office's notes are silent on an item, the AI should leave it off, which is why the prompt forbids additions.

A failed-inspection example

Say an inspection fails because a label was missing at the inverter. The crew lead tells the coordinator, who pastes the failure note and the fix into the model and asks for a short re-inspection request: what failed, what was done, when the site is ready. The reply is two sentences. The cost of that failure, in this illustrative case, is a four-day delay and a second site visit, so adding "labels match sheet E-3" to the checklist for every future job pays back quickly. Feed such lessons back into the list every month so it reflects the failures you actually have.

Adding a battery or a service upgrade mid-job

Scope changes are where narratives and checklists drift out of line with the plans. When a customer adds a battery or the electrician finds that the main panel must be upgraded, three documents change at once: the plan set (designer), the narrative and the checklist (AI-assisted). Handle it as a controlled revision. The designer issues the new revision; the coordinator updates the fact sheet and its revision line; the narrative and checklist are regenerated from the new fact sheet, not edited by hand. Then compare the old and new narratives line by line, so every difference traces to a known change.

A useful prompt for that comparison: paste the two narratives and ask the model to list every factual difference between them as a table (item, old value, new value) and to flag any sentence that changed without a fact change. The second part catches the model's habit of rephrasing for no reason, which can quietly alter the meaning of a sentence. Show the table to the designer, who confirms each difference is intended. Revision control is dull work, and it is the part that prevents a submitted narrative describing a system you are no longer building.

The grid-connection application is a separate document

Many installers fold the utility or network operator's connection application into the same admin batch, and mistakes follow because it has its own form, facts and timing. Treat it as a second package. The same fact sheet feeds it, but the checklist must come from the operator's own requirements, pasted in as before. Do not assume the permit narrative will satisfy it. Ask the model to produce a table of fields the operator's form asks for, each with the value from the fact sheet or [MISSING], and have the designer check the electrical entries. Some operators ask for equipment certification details and settings, and these come from data sheets, not from AI.

Keep the two packages' dates in one tracker so that the permit approval, the operator's approval and the crew booking line up. If your pipeline starts earlier, with leads and survey bookings, solar lead generation and fast follow-up shows how that front end feeds the permit queue.

What the qualified designer must verify

Treat this as a sign-off list. The designer, not the coordinator, initials each item before the package is filed.

1. Every number in the narrative matches the final design file: module count, DC size, inverter rating, battery capacity.
2. Plan set and calculations were produced or approved by the designer, are on the current revision and match the narrative.
3. Equipment on the data sheets matches equipment on the drawings, model for model.
4. Code and standard references in any document are checked against the edition the local authority currently adopts, and none were added by AI without a source.
5. Structural statements are supported by the roof assessment or the engineer's letter, not by assumption.
6. Grid-connection requirements, if the utility asks for a separate application, are met by a separate document and not assumed covered by the permit.
7. Anything marked MISSING or UNKNOWN in the AI's output is resolved, not deleted.

The most important of these is the last. A model told to flag gaps will flag them, and the temptation is to tidy the text and move on. A flagged gap is the AI telling you where a reviewer will push back.

Five ways permit drafting with AI goes wrong

The narrative describes last week's design

The coordinator pasted an old fact sheet after a late module swap. The narrative says 28 modules; the plans show 26. Cure: generate the fact sheet from the design export on the day you draft, and put the design revision number in the first line.

The AI recites a rule from memory

Ask a model what a permit "requires" and it will answer fluently from rules that may belong to another area or an older edition. Cure: never ask for rules; paste the office's own text.

The checklist passes a package that is missing a sheet

The AI matched the requirement "site plan" to a file called "Site-plan-draft". It was a draft. Cure: add file revision status to the file list, and read the checklist's "Unclear" rows personally.

The reply to a comment overpromises

The model wrote "all work will comply with current regulations", a statement the designer never made. Cure: replies state what changed and where, nothing more.

Customer data leaks into the prompt

Someone pastes the whole signed proposal, with the customer's name, address and bill. Cure: the fact sheet holds system facts only, and the template lists forbidden fields.

Making it a routine

Keep three files in your project system: the fact-sheet template, the three prompts (narrative, checklist, comment reply) and the designer's sign-off list. Keep a version date on the prompts and update them when an office changes its form. Each month, count how many packages were returned incomplete and why. If the number is falling, the routine works. If a rejection traces back to AI-drafted text, change the prompt or the check rather than blaming the tool.

The result is modest and useful: less coordinator time per package, fewer rejections for missing items, and a clear record of who verified what. The plan set stays with the person qualified to sign it, and so does the responsibility.

Questions installers ask about AI and permit packages

Can AI produce the plan set for a solar permit?

It should not be relied on for the plan set. Single-line diagrams, structural information, string sizing and load calculations belong to the qualified designer or engineer, who signs for them. AI can draft the narrative, cover letter and checklist around those documents, and can compare the package against the permit office's published requirements.

Which permit documents is AI good at?

Text-heavy admin: a project description, a scope narrative, a cover letter, a completeness checklist, a response to a plan-check comment and an inspection-readiness list. Each draws on facts you supply from the design, so the draft is only as accurate as what you paste in.

Should I upload a customer's utility bill or address to an AI tool?

Avoid it. Permit narratives need system facts, not personal data. Use a project code and placeholders for names and addresses, and fill the real ones in your own project system. Check your plan's privacy settings and your customer agreement before uploading anything personal.

How do I stop the AI inventing a code requirement?

Never ask it to recall the rules. Paste the permit office's published checklist and your design facts, tell it to use only those, and to mark anything it cannot support as UNKNOWN. Then have the designer confirm every reference to a code or standard against the current adopted edition.

Can AI answer plan-check comments?

It can draft the reply wording and organise a response table, but the technical answer comes from the designer. Paste the comment, the relevant design facts and what changed, and ask for a short, polite, specific reply that cites the sheet or page of the revised plans.

Further reads

• Electrical Contractor AI: Setting Up Permit and Inspection Tracking — Set up a permit and inspection tracker with AI.
• Solar Site Survey Checklist: Photos and Notes Sorted Into a Report With AI — Survey photos and notes that feed the fact sheet.
• Can AI Design a Solar Array? What Design Software Does and Doesn't Do — What design software does before the permit stage.
• Solar Installation Quality Checklist: Sort Photo Evidence With AI — Photo evidence for inspections and the closing file.
• AI for Solar Installation: Five Jobs a Small Installer Can Automate — Other admin jobs worth automating in an install firm.
• How Can Architects Use AI on Planning Applications? Where to Draw the Line — Where architects draw the AI line on applications.
• Solar Proposal Software: Adding AI-Written Cover Letters and FAQs — A worked proposal showing how to draft cover letters and FAQs with AI from your own modelled figures, without savings or green-energy promises.
• Solar Battery Quote Template: Options Explained With AI, No Backup Promises — A solar battery quote template that explains kWh against kW, lists backed-up circuits, states warranty limits and keeps AI to the installer's own numbers.
• AI for Electrical Code: Can ChatGPT Answer Code Questions Safely? — How an electrical firm can test an AI on ten known code questions, catch edition and amendment errors, and confirm each answer in the adopted code book.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general permit-package practice for small solar installers; ChatGPT and Claude plan pages (privacy and project features). Check your local authority's own published requirements.