Thermal Imaging Survey Report: Turn Scans Into Plain-English Findings

Write each finding as an observation, not a diagnosis: caption every image with where, what the colours show, the conditions and settings used, and what could explain it; state a limits paragraph once; and keep a follow-up check beside each finding. AI can draft the wording from your structured notes, but the surveyor reviews every line.

A thermal camera shows surface temperature differences, not what lies behind the surface. FLIR's own reference documentation on thermography warns that studs, water pipes and concrete columns can look like an insulation deficiency, and that moisture appears differently with the material and the time of day, so other methods should be used as well. A report that says "wet insulation" from a colour patch claims more than the scan supports. One that says "cooler band consistent with reduced insulation or air movement; not confirmed" does not.

Follow me on Instagram@sagnikteaches

Observation, likely cause, next check: a three-part finding

Most overclaiming happens in the jump from picture to cause. Take the jump out of the caption by splitting every finding into three parts, in this order. The observation says what the image shows, in neutral words, with the scale. The interpretation says what could explain it, using the hedge that matches your evidence. The next check says what would confirm or rule it out.

Connect on LinkedInSagnik Bhattacharya

The same discipline applies to single words. A heading such as "Missing insulation" is a claim, so a finding's heading should name the place and the observation instead.

Subscribe on YouTube@sagnikteaches

Record the conditions before the first image

A thermal image means little without the conditions it was taken in, so log them before you start and again if they change. FLIR's reference documentation says that when inspecting a building the temperature difference between inside and outside should be at least 10 °C (18 °F), and it names emissivity as the most important object setting to get right. It also lists reflected apparent temperature, distance and relative humidity as settings, and says the humidity can normally stay at 50 per cent for short distances and normal conditions.

Two more points from the same document belong in your routine. For draught checks, the house should be under slight negative pressure: close doors, windows and ventilation ducts and run the kitchen extractor for a while before imaging, and remember that underfloor heating can hide a draught. And an automatic scale picks the coldest and warmest points in each image as its limits, so two images of the same wall can look different. Say in every caption which scale mode you used.

Here is a filled-in conditions block for an illustrative winter evening survey of a three-bedroom house, taken after dark so that sun does not warm the surfaces. The values are examples, not targets.

Captions that carry the evidence, before and after

Here is the caption most reports carry, for an image of a bedroom wall and ceiling junction:

Image 14. Cold bridge. Missing insulation causing heat loss and likely damp. Serious defect.

Every phrase in it is a conclusion, and none can be checked. Now the same image captioned so that a reader can see what was measured and what was inferred:

Image 14 (paired with photo 14V). Bedroom 2, north external wall, ceiling junction. A band along the junction reads about 2 °C below the wall face 600 mm lower, on a manual scale of 14.8 to 19.6 °C (iron palette). Settings: emissivity 0.94, reflected apparent temperature 19.0 °C, distance 2.5 m, relative humidity 50 per cent. The pattern is consistent with reduced insulation or air movement at the junction; timber framing can produce a similar pattern, so this is not confirmed. Suggested check: borescope inspection at two points, or a repeat scan after a cold night.

That caption has seven working parts. The image and its paired visible-light photo, the location, the observation with the scale, the settings, the hedged interpretation, an explicit "not confirmed" and the suggested check. Build every caption from the same seven, so a reader learns where to look.

Two habits keep the numbers honest. First, name files so that an image, its visible photo and its notes row share one ID (014, 014V and row 14), so nobody has to guess which picture a caption belongs to. Second, call spot readings apparent temperatures. FLIR's glossary defines apparent temperature as the uncompensated reading, containing all the radiation that reaches the camera whatever its source, so a spot value means little without the emissivity and reflected temperature that shaped it. A comparison between two areas of the same material, in the same image and under the same settings, leans less on those settings than a single absolute figure does, which is why the example caption reports a difference.

The limits statement: write it once, put it up front

The limits paragraph tells the reader what a thermal survey cannot do, and it protects you only if it is accurate and visible. Put it near the front, not in an appendix. Cover the scope and what was not accessible, the method and its blind spots, the conditions, the fact that the survey is a snapshot, who may rely on it, what the report is not, and who reviewed it. An illustration of the wording:

This is a thermal imaging survey of the accessible surfaces listed in Section 2, carried out on [date] in the conditions recorded in Section 3. Thermal imaging records the apparent temperature of surfaces. It cannot see through materials and cannot measure moisture content, insulation thickness or structural condition. Patterns are interpreted in context and are not conclusive; findings marked "not confirmed" need a second method. The survey is a snapshot of conditions on the day. Areas that were not accessible or could not be assessed are listed in Section 2. This report is for [client name] and should not be relied on by anyone else. It is not a structural, compliance or valuation report. Findings were reviewed by [surveyor name and qualification] on [date].

Do not let an AI assistant write or edit this paragraph. Use wording your professional body or insurer has approved, or have an adviser write it once and paste it in every time. Check what your insurer expects before you adopt any AI-assisted routine at all.

A prompt that turns structured notes into captions

Give the assistant one row per image, written by you on site or from the camera software. Include the settings in the row, since the model cannot see them. A row looks like this:

ID 14 | Paired photo 14V | Location: Bedroom 2, north external wall, ceiling junction
Settings: emissivity 0.94, reflected 19.0 C, distance 2.5 m, RH 50%, manual scale 14.8 to 19.6 C, iron palette
Observation: linear cooler band along junction, about 2 C below wall face 600 mm lower
Surveyor's reading: possible reduced insulation or air movement; timber framing could give a similar pattern
Next check: borescope at two points; repeat scan after a cold night

You are drafting captions for a thermal imaging survey report. The surveyor will review every caption.
Use ONLY the facts in each NOTES row. Do not add temperatures, causes, building details or recommendations.
For each row write one caption of 60 to 90 words in this order:
1. Image and paired photo IDs, and the location.
2. What the image shows, with the scale and palette, in neutral wording ("reads about X below"). Never "is".
3. The settings, in one sentence.
4. The interpretation, using only "consistent with", "may indicate" or "could not be assessed", taken from the surveyor's reading.
5. "Not confirmed" if no second method is listed, then the suggested check from the row.
Never use: defect, failure, leak, damp, wet, dangerous, serious, poor workmanship, non-compliant, replace.
If a row lacks a field, write [missing: field name] and do not guess.
NOTES
[paste rows]

When a row is thin, an assistant fills the gap with confident invention. A first draft of that kind, as an illustration:

Image 14. The ceiling junction in Bedroom 2 is losing heat through missing insulation, with a 2.3 °C drop in temperature that suggests a moisture problem. Insulation should be topped up. The camera was set to standard settings.

Five problems hide in those three sentences. It states a cause ("missing insulation") where your row said "possible". It invents a precise figure, 2.3 °C, where the row said about 2. It names moisture, which your row never mentioned. It adds a remedy ("topped up") that nobody asked for. And "standard settings" replaces the actual emissivity, reflected temperature and distance, which is the one thing the model could not know. Paste the five problems back as a correction and rerun it from the row alone. The rewritten caption should then match the good caption in the previous section almost word for word; if it does not, the row is probably missing a field.

A plain-English summary for the client, hedges intact

The client reads the first page and the summary, and rarely the caption pages. Write the summary last, from the finished findings register, in short sentences, putting the findings in the order you would act on them. Technical language needs translating, but the hedges must survive translation. Compare:

Technical: Thermal anomaly at the eaves junction, 2 K differential, indicative of insulation discontinuity or air ingress.

Plain: At the top of the north bedroom wall, the surface is slightly cooler than the wall below it. That can happen when insulation is thin or missing at that point, or when cold air gets in. The roof timbers behind the plasterboard could cause a similar pattern. We have not confirmed the cause. We suggest checking through two small inspection holes.

Rewrite each finding below for a homeowner. Keep every hedge exactly as written (may, could, consistent with,
not confirmed, could not be assessed). Do not add causes, remedies, costs or urgency. Do not use the words defect,
failure, leak, damp or dangerous. Use short sentences. For each finding give: what we saw, what it might mean,
what we suggest next. Then list any finding whose wording you changed so that I can check it.
FINDINGS
[paste the reviewed findings register]

The last line of that prompt matters most. It makes the assistant show its changes, so the reviewer reads those lines first and does not hunt for them.

A worked survey: from 112 images to a findings register

Take an illustrative survey of a three-bedroom house for an owner who asked why the upstairs is cold. The surveyor captures 112 images over about 90 minutes, keeps 38 that show something, pairs each with a visible-light photo, and ends with 12 findings. The time budget below is my estimate for that job, so adjust it for yours.

Added up, that is 305 minutes, a little over five hours from arrival to a reviewed register, and the assistant accounts for 10 of them. The drafting step is short. The saving comes from not typing 38 captions from scratch, and the review step is the one that cannot shrink. A slice of the finished register shows the shape:

The register is also where severity ordering lives. If you already use a severity scheme from snagging or inspection work, the snagging report template shows one, and building a comment library shows how to keep approved wording for findings you write often.

Review before sign-off: eight checks

1. Every caption has its conditions and settings, or points to a conditions block that applies.
2. Every temperature in a caption appears in the image file or the notes row, with the same rounding.
3. The scale and palette are stated, and the same image shows them.
4. Each interpretation uses a hedge that matches its evidence, and "confirmed" appears only where a named second method exists.
5. Each finding has a next check or an explicit statement that none is suggested.
6. No caption or summary says more than the scope: no blame, no cost, no remedy, no safety opinion.
7. The limits paragraph is present, near the front, and was not edited by an assistant.
8. The reviewing surveyor's name, qualification and the review date are on the report.

A final test costs five minutes. Give the summary page to someone who was not on site and ask them to say, in their own words, what was found and what happens next. If they say anything stronger than your hedges, the summary needs another pass. The liability side of this routine is covered in the tutorial on AI report narratives and liability checks, and for a closely related damp question, the damp survey report tutorial shows how findings turn into quote lines.

Three situations that change what you can say

The day fell short of your minimum. If the inside-to-outside difference was well under your usual minimum, a weak pattern could mean a small problem or just a small temperature difference. Say that the survey was limited by conditions, list what could not be assessed, and offer a rescan on a colder evening. A thin image set dressed up as a full survey is the riskiest report you can write.

The surface is reflective. Glass, polished metal and shiny finishes have low emissivity and mirror whatever is in front of them, including you. FLIR's document says emissivity normally runs from about 0.1 to 0.95, with a highly polished mirror surface below 0.1. Mark such surfaces as not assessed, or as limited, rather than quoting a temperature.

The client asks "is there damp?" The scan alone cannot answer, and it is better to say so kindly than to say yes or no. A reply that works: "The scan shows a cooler area on that wall that is consistent with moisture, but also with missing insulation or a cold spot in the structure. To find out, we would measure the wall with a moisture meter and, if needed, open a small inspection hole." The tutorial on drain camera survey reports shows the same hedged style for another camera-based trade.

Which files and details stay out of the chat

Thermal surveys come with a client's address, floor plans and photographs of the inside of a home. Decide what an AI tool may see before you start. Notes rows with a job number and room names are usually enough for captions, so leave names, addresses and contact details out. If you do upload an image, remove location data first, keep the original radiometric files in your own storage, and treat the assistant's description of an image as a colour description only. It cannot tell a 24 °C patch from a 31 °C one unless the scale is legible and correct, and it does not know the emissivity you set. Use it for wording, never for diagnosis.

Questions surveyors ask about writing up scans

Can I upload thermal images to an AI assistant?

You can, but an assistant sees colours, not temperatures, unless the scale is readable, and it does not know your emissivity or reflected temperature setting. It is safer to give it your structured notes and let it draft wording. If you do upload images, strip the address and location data first and never ask it to diagnose.

Should the report give temperatures?

Give apparent temperatures only with the scale and the settings beside them, and lead with relative differences, such as a band reading about 2 degrees below the wall face. Absolute readings depend on emissivity, reflected temperature and distance, so a bare number without those settings invites a challenge.

How do I word a finding I am unsure about?

Use the hedge that matches the evidence. Say reads or shows for what the image displays, consistent with or may indicate for a pattern with several possible causes, confirmed by a named method only when a second check agrees, and could not be assessed where access or conditions stopped you.

Does AI replace the surveyor's sign-off?

No. It drafts captions and summaries from your notes, and the surveyor remains accountable for every line. Put the reviewer's name and the date on the report, and check your insurer's and professional body's requirements for AI-assisted drafting before you adopt the routine.

Further reads

• Mould Inspection Report Template: Findings, Photos and Limits With AI — See how findings, photos and limits fit another inspection report.
• Roof Inspection Reports: Photos to a Customer-Ready Report With AI — Turn roof photos into a customer-ready report.
• Chimney Inspection Report: Can AI Turn Camera Findings Into Plain English? — Plain-English wording for camera findings in another trade.
• Structural Engineer Inspection Report: Photos and Notes to a Draft With AI — Draft from photos and notes where an engineer signs off.
• How Surveyors Use AI to Turn Site Notes Into Reports — Start from site notes in the wider surveying workflow.
• Can Home Inspectors Use AI Without Taking On Extra Liability? — Decide which report jobs are safe to hand to AI.
• 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.
• Sewer Camera Inspection Reports: From Defect List to Customer Summary — How a drain survey firm turns coded camera findings into a customer summary with AI, with a prompt, a worked defect list and the surveyor's checks.
• Solar Site Survey Checklist: Photos and Notes Sorted Into a Report With AI — Grouped photo checklist with why and how to verify each item, then a tagging system and prompts that sort photos and notes into a survey report.
• 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.
• Schedule of Condition Template: Draft From Photos With AI, Then Verify — A fill-in schedule of condition template, plus the steps to draft entries from site photos with AI and verify every line on site.
• Use of AI in Building Surveying: Client Data and UK GDPR Basics — A yes/no table for client details and site photos, what to redact, which plan settings matter and what processing terms to look for. Not legal advice.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: FLIR reference documentation on thermography (emissivity, reflected apparent temperature, level and span, palettes, building inspection notes). Survey figures in the examples are illustrations.