Mould Inspection Report Template: Findings, Photos and Limits With AI

Write the report from three inputs only: your dated site notes, your numbered photos and any lab results. Put them into a fixed skeleton (scope, method, findings by area, photo log, limitations, next steps), let AI turn notes into plain sentences and captions, then trace every statement back to its source before you sign.

AI is a drafting assistant here, not a witness. It cannot tell you whether a dark stain is microbial growth, it can misread the digits on a moisture-meter display, and it will turn "possible" into "confirmed" if you let it. So the notes stay yours, the readings stay yours, and anything about health stays out of the report altogether.

Follow me on Instagram@sagnikteaches

The rest of this follows one sole inspector through one booking: a two-storey rental where the tenant reported a musty bathroom. The inspector spent 70 minutes on site, took 41 photos, logged nine moisture readings and sent two tape-lift samples to a lab. Those figures are illustrative. The workflow is the part to copy.

Connect on LinkedInSagnik Bhattacharya

You may see the subject spelt "mold" in report software and in searches; the structure is identical. If your licensing body or your standard prescribes headings or defined terms, use those, and treat the skeleton below as the fallback.

Subscribe on YouTube@sagnikteaches

The skeleton: ten parts in the order a client needs them

Most mould reports run long because the writer follows the order of the site visit. Clients read in a different order: what did you find, how sure are you, what happens next. The skeleton puts the answer first and the evidence behind it.

1. Job details. Report number, client, a job code in your working files instead of the full address, date and time on site, weather, who attended and who instructed you.
2. Scope. What you were asked to look at, and what you were not ("bathroom and bedroom 2 only").
3. Method and instruments. Visual inspection, the moisture meter's make, model and scale (relative or percentage), thermal camera if used, how any samples were taken, and the dates your instruments were last checked.
4. Summary. Five lines at most, in plain words: what was seen, what was measured, what is not yet known and what you recommend. Clients often read only this part, so write it last.
5. Findings by area. One block per room or building element: observation, readings, photo numbers.
6. Samples and lab results. Sample IDs, where each was taken, and the lab's own report attached unaltered.
7. Probable moisture sources. What you think is feeding the damp, how sure you are, and what is not confirmed.
8. Recommendations. Fix the moisture first, further investigation, which specialists, who should carry out any remediation. No health advice.
9. Limitations. What you could not see, reach or test, and what the report cannot say.
10. Appendices. Photo log, a floor plan with photo positions marked, the lab certificate, instrument details.

AI earns its keep in parts 5 and 10 (the findings and the photo log), gives you a first draft of part 9, and helps with the wording of parts 4 and 8. It should not touch part 6. The lab's numbers go in exactly as the lab wrote them, and the lab's report goes in as an attachment, never as a retyped summary.

The raw notes: what the inspector typed on the Tuesday

These are the notes from the rental job, typed into a phone while walking round. They are untidy on purpose. What matters is what they contain.

JOB 0418 | Tue 14:10 | 2-storey rental house | tenant present | dry, 9C outside
Instruments: pin/pinless moisture meter (relative scale), thermal camera

BATH (1st fl): ceiling above shower, dark stain approx 40x30cm, plaster feels damp.
Meter 28 (rel) at stain, 11 on ceiling by door. Extractor fan runs, duct not seen
(loft hatch painted shut, not accessed). Black marks on grout line at shower screen
seal, look like surface mould. Photos 5-12.
BED 2: external wall behind wardrobe, spotting on bottom 50cm, wardrobe touching wall.
Meter 19 at spotting, 10 on wall beside. Photos 18-24.
KITCHEN: nothing visible. Meter 9-11 all walls, under sink dry. Photos 30-31.
LANDING: hairline crack in ceiling, no staining. Photo 33.
SAMPLES: tape lift BATH ceiling = S1, tape lift BED2 wall = S2. To lab Wed.
NOT SEEN: loft, bed 1 (locked; tenant says owner's storage), behind kitchen units.

Three habits make notes safe to hand to an AI.

• Photo numbers sit next to the thing they show. "Photos 5-12" at the end of the bathroom line means nobody has to guess which picture belongs to which finding.
• Every reading carries its scale and its comparison. "28 (rel) at stain, 11 on ceiling by door" tells a reader that 28 only means something next to 11.
• There is a NOT SEEN line. The loft, the locked room and the space behind the units are written down on the day, so the limitations section comes from the notes instead of from memory.

If you dictate rather than type, the same habits apply. The tutorial on surveying site notes and photos covers turning rough spoken notes into a draft report.

Turning notes into findings without letting AI add anything

The prompt below does most of its work in the rules. Each rule exists because of a specific way a drafting assistant goes wrong: it rounds, converts, upgrades certainty or fills a gap with plausible detail. Paste the notes underneath it, in a business plan or a chat with training switched off (more on that further down).

You are drafting the "Findings by area" section of a mould inspection report. I will edit it and sign it. Use British English.

Rules:
1. Use only the notes below. If something is not in the notes, write "not recorded".
2. Keep every number, unit, scale and photo reference exactly as written. Do not convert, round or estimate.
3. Keep my certainty words. "Appears", "possible" and "look like" stay as written. Never upgrade them to "confirmed", "caused by" or "due to".
4. Do not say what dark staining or marks are. Call them "dark staining" or "dark marks" unless my notes say "visible growth".
5. No health statements. No "safe", "unsafe", "toxic" or "mould-free".
6. One short paragraph per area, then "Photos:" and the numbers.
7. After the draft, list anything in my notes you could not place.

NOTES:
[paste notes]

Here is the kind of first draft that comes back. It is illustrative, and the wording varies from run to run.

Bathroom (first floor). Dark staining about 40 x 30 cm on the ceiling above the shower; the plaster feels damp. Moisture meter reading 28 on the relative scale at the staining and 11 on the ceiling by the door. The extractor fan runs; the duct was not seen because the loft was not accessed. Dark marks on the grout line at the shower screen seal. Photos: 5 to 12.

Bedroom 2. Spotting on the lower 50 cm of the external wall behind the wardrobe, which touches the wall. Readings of 19 at the spotting and 10 on the adjacent wall. Photos: 18 to 24.

Kitchen. No visible staining. Readings of 9 to 11 on all walls; the area under the sink is dry. Photos: 30 and 31.

Could not place: landing ceiling crack (photo 33); samples S1 and S2 (these belong in the samples section); the NOT SEEN line (this belongs under limitations).

It is close, and close is the danger. A draft like this still needs four edits, all of them typical.

• The inspector's note said the marks on the grout "look like surface mould". The draft softened that to "dark marks", which is safe. Whether to write "appear to be surface growth" is the inspector's call. The prompt protects against upgrading; it cannot make the judgement for you.
• Some runs add a percent sign ("28%"). The meter's scale was relative, so the number stands alone, followed by "on the relative scale".
• "The extractor fan runs" can come back as "the extractor fan is working correctly". Running is not the same as extracting adequately, so the original wording goes back in.
• The "could not place" list correctly flags the landing crack, the samples and the NOT SEEN line. Move each to its own section. Ignoring that list is how items quietly disappear from a report.

Photo captions that describe, locate and measure

A caption has one job: let a reader who was not there find the spot and see what you saw. It is not the place for diagnosis. The pattern that works is the same every time: photo number, location, what is visible, scale or reading.

When you give photos to an AI, type the room and surface into the message. OpenAI's developer guide says file names and metadata are not read, so "0418_BATH_07.jpg" tells the model nothing unless the same words appear in your prompt. The assistants also cap what you can send at once. As of October 2026, ChatGPT takes images up to 20 MB each, Claude takes 20 images per message at up to 10 MB each, and Gemini takes 10 files per prompt. One room at a time, with eight or ten photos and that room's notes, stays inside all three limits and stops captions from bleeding between rooms.

Write a caption for each photo in the bathroom set. Format: "Photo [number]. [Location]. [What is visible]. [Scale or reading, if I gave one]."

Rules: Describe only what is visible, plus the notes I have pasted. Do not name what a stain or mark is. Do not estimate sizes from the picture; use my measurements or write "no measurement taken". If a photo shows a meter display, do not read the digits; write "meter display, reading per notes". If you cannot tell what a surface or material is, write "material not identifiable from photo".

Photo list (number | location | note):
5 | bathroom, wide shot from door | whole ceiling and shower
7 | bathroom, ceiling above shower tray, looking up | stain, tape measure at 40 cm
8 | same, close up | meter on stain, reading 28 rel
9 | bathroom, ceiling by door | meter, reading 11 rel

A good first pass looks like "Photo 7. Bathroom, ceiling above the shower tray, looking up. Dark staining with defined edges; tape measure in frame at 40 cm." The failures are just as predictable. A caption sometimes names the stain as mould, estimates "about 60 x 40 cm" from the picture although the tape in frame says 40, or reads a display aloud with one digit wrong. Each one is a rule the model broke, and the fix is always to edit the caption back to your notes.

That last failure deserves its own habit. Vision models can misread small displays, and a wrong digit in a caption is the sort of error a client's expert finds first. Keep the photo of the meter as evidence, type the reading in your notes, and let the notes supply every number. The finished photo log then becomes a plain table, one row per picture.

The same caption method carries over to other trades. The roof inspection photo report tutorial shows it applied to a roof, with the customer-facing layer on top.

Recommendations: let AI pick from your list, not write new advice

Recommendations carry the most weight if a report is ever disputed, so this is the one section where a blank-page draft is the wrong move. Keep a short library of your own standard recommendations, each in wording you would say out loud, and ask the AI to choose and adapt.

Below are my standard recommendations (R1 to R6) and my findings. Choose the ones that apply, put them in priority order, and adapt the wording to name the specific room. Do not add any recommendation that is not in my list. Do not give health advice. If a finding has no matching recommendation, tell me instead of inventing one.

R1 Resolve the moisture source before any cleaning or remediation.
R2 Arrange access to [area not inspected] and a follow-up inspection.
R3 Have the extractor fan and duct checked by a suitable tradesperson.
R4 Consider further investigation of [location], including invasive or specialist testing.
R5 Move furniture away from external walls so air can circulate.
R6 Any remediation should follow a written scope from a suitably qualified contractor.

FINDINGS:
[paste the findings section]

The final instruction is the valuable one. It turns the model's habit of filling gaps into a flag for you. On the rental job, a typical answer picks R1, R3, R2 for the loft and bedroom 1, R5 and R6, and adds a note: "Finding: hairline ceiling crack on landing, no matching recommendation." The inspector decides whether that crack needs a line of its own or a mention under other observations. Either way, a person deals with it.

The limitations paragraph: a standing base, then the job's own gaps

Every report needs a base paragraph that does not change between jobs, and a short job-specific list that always does. The base lives in your template. The list is written from the NOT SEEN line.

This report is based on a visual inspection of the areas listed in the scope on [date], together with the readings and samples described under Method. It records conditions on that day only. Moisture meter readings are indicative: they are compared with readings on nearby materials and do not identify a moisture source on their own. Sample results describe the sampled location on the day of sampling. This report does not assess health effects and must not be used for that purpose. Concealed areas, areas behind furniture or fixed fittings, and areas that could not be accessed on the day were not inspected unless stated below.

The job-specific part is where a drafting assistant is useful, because it is a rewrite of three lines of notes into three clear sentences. Paste the NOT SEEN line and ask for one sentence per item, each with its reason, no reassurance and nothing added. For the rental job a good result reads like this:

• The loft and the extractor duct were not inspected: the loft hatch was painted shut.
• Bedroom 1 was not inspected: it was locked, and the tenant stated that it holds the owner's storage.
• The space behind the kitchen units was not inspected.

Two cautions. First, a limitation does not cover what you could reasonably have checked. If the loft is the likely source, say so under recommendations and offer a return visit instead of hiding behind the hatch. Second, boilerplate that never changes tells a reader nothing about this property. Whether your standing wording is enough for your insurer is a question for the insurer or a lawyer, not for an AI.

Medical and unproven claims: what to cut and what to write instead

A mould report describes a building. It does not describe anyone's health, and it should not claim more than the evidence shows. Drafting assistants lean towards dramatic, confident phrasing, so the prompt bans the worst of it and you search the draft for the rest.

Put the left-hand column into the prompt as a banned list, then search the finished draft yourself for: toxic, safe, unsafe, caused, confirmed, mould-free, health, illness. Each hit is either a claim you can support from the notes or a sentence to rewrite. If a client asks about symptoms, the honest reply is that you cannot assess health effects and they should speak to a doctor.

The check before you sign: every statement traced to its source

The inspector signs, so the inspector checks. A workable method is to go through the draft with a highlighter and mark three kinds of sentence: anything with a number, anything with a certainty word (appears, confirmed, caused, likely), and anything telling the client what to do. Then trace each one.

A second chat can help with a final sweep. Paste the notes and the finished report into a fresh conversation and ask it to list every sentence the notes do not support. It will catch obvious slips and it will also raise false alarms. Treat it as a second pair of eyes after your own check, never instead of it. The tutorial on liability checks for AI-written home inspection narratives goes further on what to confirm before a client receives anything. If you keep a library of standard comments, the comment library tutorial shows how to build one, so the AI picks from your wording instead of inventing its own.

A second job: an empty flat with no sampling

The skeleton bends. A landlord asks for a check on an empty flat before new tenants move in: all rooms, no sampling, 45 minutes on site. Part 6 shrinks to one line ("No samples were taken"), the scope covers every room, and the summary matters most because the landlord may read nothing else. The meter read between 8 and 12 on the relative scale in every room.

This is where the pull towards overstating is strongest, because there is good news to report. A first draft might say "The flat is free of mould and safe to rent out". The notes support something narrower.

No visible growth was observed in the areas accessed on [date]. Moisture meter readings were between 8 and 12 on the relative scale in all rooms. The areas listed under Limitations were not inspected.

Same facts, no promise. The limitations list is shorter but it still exists: the rear wall of the fitted wardrobe, the boxed-in pipework in the bathroom and the loft. If a damp problem does appear in that flat later, the report should still read as true. When a check between tenants turns into a repair quote, the damp survey report tutorial covers writing findings and quote lines together.

What goes into the AI: client details, photos and location data

A mould report carries a client's name, an address and photos of the inside of their home, sometimes with belongings, post and family photographs in shot. Three habits cover most of the risk.

• Use the job code, not the address, in prompts and file names. The drafting step does not need to know where the building is.
• Crop or skip photos showing people, documents, mail and anything personal. A caption can say "wide view of the room" without that picture going to an AI at all.
• Remove location data. Phone cameras usually save where a photo was taken inside the file. OpenAI says file names and metadata are not read by its models, but the file you upload is still a copy of your original, and the same file may later land in a shared drive or a PDF. Turn location off for the camera app on inspection days, or strip the data before files leave your device.

Which plan you use matters as much. Business plans (ChatGPT Business, Claude Team) do not train on your content by default. Consumer plans have a switch in privacy settings that you should check is off, because defaults change. Both business plans need at least two seats, so a one-person firm pays for two (about $40 to $50 a month at list prices) or uses a personal plan with training switched off. Temporary Chat in ChatGPT is not used for training, though a copy may be kept for up to 30 days. Finally, read your own instruction letter and privacy notice. If you promise clients that their photos stay in-house, an upload to a third-party AI may break that promise, and the wording needs updating before the workflow does.

The time saved on one report, and what the AI costs

For the rental job, the inspector timed each step against a comparable report written by hand. Treat the figures as illustrative and time your own next three reports before believing any of it.

The saving is 55 minutes a report, and the checking step grows rather than shrinks, which is how it should be. At 16 reports a month that is about 15 hours (16 x 55 = 880 minutes). A personal plan costs about $20 a month at list price: ChatGPT Plus is $20 and Claude Pro is $20, or $17 a month on annual billing. A two-seat business plan costs $40 to $50. The software cost is trivial. The real costs are an afternoon building the skeleton and prompts once, and the discipline of never skipping the check.

Questions mould inspectors ask about AI-drafted reports

Can AI identify mould from a photo?

Not reliably, and a report should not claim it can. A photo cannot separate microbial growth from soot, dirt or a stained coating, so identification stays with the inspector and, where you sample, the lab. Use AI to describe what is visible: colour, size, position, and the reading you took at that spot.

Should the report say a property is safe to live in?

Only if your licence or standard requires that judgement and you are qualified to make it. Otherwise leave it out, describe what you found and what you could not inspect, and tell the client to take health questions to a doctor. Ask your insurer or a lawyer which wording your own reports should use.

Can AI write the recommendations section?

It can choose from a library of recommendations you have written and tidy the wording, which is safe. Letting it invent advice is not, because recommendations carry the most weight if there is a dispute. Give it your standard list, tell it to add nothing, and delete any line you would not say to the client in person.

Do I keep the AI chat with the job file?

The record is your notes, photos, lab report and the signed PDF. The AI draft is scaffolding. Delete chats on your normal schedule, keep nothing in them that you would not keep in the job file, and never treat a chat as evidence of what you saw on site.

Further reads

• Structural Engineer Inspection Report: Photos and Notes to a Draft With AI — The same notes-and-photos method for a structural inspection.
• Water Damage Restoration Software: Moisture Logs and Scope Notes Checklist — Moisture logs and scope notes for the restoration side of a damp job.
• Septic Inspection Report: Can AI Write It From Field Notes? — Another field inspection turned into a report from rough notes.
• Chimney Inspection Report: Can AI Turn Camera Findings Into Plain English? — Plain-English findings from camera evidence.
• Sewer Camera Inspection Reports: From Defect List to Customer Summary — From a defect list to a customer summary.
• Fire Door Inspection Report: AI Drafting and What You Must Record — What you must record when AI formats the report.
• Water Damage Restoration Leads After a Flood: Answer and Book With AI — A six-question intake script, storm-night priority tiers, overflow answering costs and an insurer-ready photo list for a water damage restoration firm.
• Architect Site Visit Report: Notes and Photos to a Clear Report With AI — A worked notes-to-report example for architects: observation, instruction and action columns, photo captions, and a rule that keeps unseen items out.
• Home Energy Audit Report: Turn Findings Into a Client Report With AI — Turn audit notes into a client report with AI: a findings table, priority order, plain-English summary and savings as ranges, with every figure checked.
• Leak Detection Reports: AI Turns Findings and Photos Into Customer Reports — A leak detection report workflow: structured field sheet, numbered photos, two AI passes, wording that avoids liability and a check on every figure.
• Snagging Report Template: Sections, Severity Order and AI Captions — A snagging report laid out in severity order, with a room-by-room defect log, a caption formula and AI caption examples the inspector checks before sending.
• Structural Engineer Report Template: A Plain-English Summary You Can Check — A blank report skeleton, a client summary box, and a sentence-by-sentence check that stops AI changing the technical meaning, shown on a sample extract.
• Thermal Imaging Survey Report: Turn Scans Into Plain-English Findings — How a thermal imaging surveyor writes captions, conditions and a limits statement, uses AI to draft from structured notes and reviews every finding.
• Can AI Help Write an Arborist Report? What Stays With the Arborist — Use AI to turn tree survey notes into a readable report draft while measurements, risk ratings and advice stay with the arborist.
• Using AI in Environmental Consulting: Draft Site Report Sections — Prepare site-report sections from controlled notes and lab results, with a traceable check of every number and conclusion.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI, Anthropic and Google help and developer pages on image and file upload limits and on how uploaded content is used, as checked in October 2026.