Schedule of Condition Template: Draft From Photos With AI, Then Verify

Yes, AI can draft a schedule of condition from site photos by writing room-by-room entries with photo references in neutral wording. It cannot confirm what it cannot see, so the surveyor must visit, check every entry against the real surface, correct or delete anything unsupported, and only then issue the schedule under their own name.

A schedule of condition is a record of the state of a property at a point in time, usually attached to a lease, a works agreement or a neighbouring-building arrangement so that later disputes about damage can be settled against a dated baseline. Its value is precision and neutrality. AI is fast at the first and needs supervision on the second. That is why the template below carries a "verified on site" column on every line: the surveyor's check is part of the document, not a step that happens somewhere off the page. Nothing here is legal advice.

Follow me on Instagram@sagnikteaches

A fill-in schedule of condition template, line by line

Copy this structure into a document or spreadsheet. Every entry carries a location, an element, a description, a photo reference and a verified flag, so the surveyor's check is visible on the page.

Connect on LinkedInSagnik Bhattacharya

Above the table, put a header block: property address, date and time of inspection, weather, who attended, who instructed the schedule, the purpose, and a scope statement ("visual inspection only; no opening-up; areas not accessed are listed below"). Below it, a limitations section lists anything not seen: locked rooms, roof void, areas behind furniture. A signature line for the surveyor closes the document.

Subscribe on YouTube@sagnikteaches

Photographing so the AI has something to work with

The quality of the draft is set at the camera. A model can describe a clear photo and will fabricate around a blurry one. Settle a repeatable routine before the visit:

• Same order every room: clockwise from the door, then ceiling, then floor. That order becomes the entry order.
• A reference object: a folding rule or a card of known size in the frame for each defect, so size can be read.
• A wide shot, then a close-up: the wide shot shows position; the close-up shows detail. Name them as a pair, such as 1-03a and 1-03b.
• Raking light: hold a torch low and across a wall to show cracks and ripples that a flat light hides.
• Limit people and papers: avoid faces, screens and documents in frame; use a grey card or blank wall behind close-ups where you can.

Keep the camera's original files with their date stamps, and name them by the template reference. Edited, cropped or filtered copies are for convenience only and should never replace the originals.

Prompting a first draft from one room's photos

Upload a room's photos and ask for entries in the template's shape. A tool that reads images, such as ChatGPT or Claude on a plan that allows image upload, is enough; what ChatGPT can and cannot read from photos explains the common failure points. Check your plan's privacy setting before uploading anything from an occupied property, and see the privacy note in the FAQ.

I am a surveyor preparing a schedule of condition. I will upload photos of one room.
For each photo, produce table rows with columns: Ref, Element, Description, Photo.

Rules:
- Describe only what is visible. Do not state causes (no "damp", "subsidence", "settlement").
- Use neutral terms: "staining", "hairline crack", "paint loss", "gap".
- Give a size only if a rule or reference object is in the photo; otherwise write
  "size not measurable from photo".
- Give position as wall (N/E/S/W as I specify), height band (low/mid/high) and
  distance from the nearest corner or opening if visible.
- If a photo is too dark, blurry or does not show the element, write "NOT CLEAR - recheck on site".
- At the end, list any element that is normally part of a room schedule but has no photo.

The last three rules matter most. The instruction to avoid causes keeps the record neutral. The "NOT CLEAR" rule turns a model's tendency to fill gaps into a recheck list. The missing-elements list catches the radiator, the window sill or the ceiling that nobody photographed.

A realistic sample output, and what you would fix

Illustrative output for a bedroom wall photo with a folding rule in frame:

1.22 | North wall | Painted plaster | Hairline crack running diagonally from upper right corner of window head, approx. 300 mm long (rule in frame). No visible displacement. | 1-22a, 1-22b
1.23 | North wall | Skirting | Gloss paint worn to bare timber at the left end, approx. 150 mm. | 1-23a
1.24 | North wall | Radiator | Rust marks at lower valve. Possible leak. | 1-24a

Entries 1.22 and 1.23 are close to usable after the surveyor confirms position and size at the wall. Entry 1.24 breaks the rule: "possible leak" is a cause or a guess. The surveyor replaces it with "Brown staining at lower valve union, approx. 20 mm across, dry to touch at inspection", but only after touching the valve and the floor below. This is a typical correction: the AI reports a conclusion the photo does not support, and the human replaces it with an observation.

Verifying every entry on site

Verification is the surveyor's job and it is not optional. The routine below keeps it systematic. Print the draft, take it to the property, and work room by room with a pen:

1. Confirm presence: does the defect or condition exist where the entry says? Delete any entry that has no match.
2. Confirm position and size: measure with the rule, not the photograph. Photos distort size by angle and lens.
3. Test, within scope: operate windows and doors, press a loose tile gently, check a surface by hand. Record "stiff", "loose" or "sticks" where it applies. Do not open up fabric unless the instruction allows it.
4. Look for the unphotographed: behind doors, under sinks, above wardrobes, ceiling corners, window reveals. Add entries with new photographs.
5. Replace opinions with observations: every cause, prediction or adjective like "bad" or "serious" becomes a measurable description or comes out.
6. Initial and date each line: the "verified" column is the audit trail. An entry without initials has not been checked and should not be issued.

Then a second person, if you have one, spot-checks ten entries at random against the property. If more than one in ten is wrong, the batch is not ready, and the prompt or photo routine needs work.

A filled-in room, start to finish

Say a surveyor is recording a first-floor bathroom ahead of building work next door. Here is a short, illustrative section, after verification:

Notice what is missing: no causes, no recommendations, no "wear and tear" judgments. A schedule of condition records. It does not diagnose, and mixing the two is how records become arguments.

Three more worked examples of AI entries and the fix

Before and after: one stained ceiling entry

Before (AI draft): "The ceiling is in poor condition with obvious damp damage." After (surveyor): "Brown staining, approx. 600 x 400 mm, centred 1 m from east wall. Paint blistered at centre. Dry to touch." The second version supports a later comparison; the first starts one.

A photo that misleads

A wide shot of a lounge shows a clean carpet. The AI writes "carpet in good condition throughout". On site the surveyor finds a worn patch behind the sofa, out of frame. The corrected entry records the worn area and its size, and the surveyor adds a line to the routine: move light furniture or list the areas not accessible. The lesson is that an AI summary of a photo describes the photo, not the room.

A quick sum for time

Say a three-bedroom house takes 280 photos and the surveyor used to spend about five hours typing entries from notes. With AI drafting, upload and prompting for ten rooms takes roughly 50 minutes, and verification on site takes the same time it always did. The saving comes only if the checks are done and the photo routine is tight. If verification is rushed, the saved typing time reappears later as a dispute. See how surveyors use AI to turn site notes into reports for the notes-first route, which suits sites where photography is restricted.

Keeping the wording neutral

A short style sheet lets the AI and the surveyor write alike. Put it at the top of every prompt:

• Use present tense: "The paint is flaking", not "was flaking".
• Give each defect a size and position, or say it could not be measured.
• Prefer "approx." with a unit over "small", "large" or "minor".
• Do not use "poor", "good" or "fair" without a defined scale in the header. If you use a scale, define it once (for instance, "good: no defects visible; fair: minor surface marks; poor: loss of material or function").
• Never state cause, age, cost or recommended work in the schedule itself. Put those in a separate report.

Neutral wording protects every party. A tenant, a landlord or an adjoining owner reading an entry should not be able to tell whose side the surveyor is on.

Writing the header and limitations block with AI

The entries get the attention, but the header and limitations decide what the schedule can be relied on for. These are safe to draft with AI because they contain no observations, only your facts. Give it the facts and ask for plain wording:

Write the header and limitations section of a schedule of condition.
Facts: inspection date [date], 10:15 to 12:40, dry and overcast. Present: surveyor
and the occupier's representative. Purpose: baseline record before neighbouring
building work. Visual inspection only, no opening-up, no testing of services.
Not accessed: loft hatch (painted shut), rear shed (locked), under-stairs cupboard
(stacked with boxes). Use short sentences. Do not add anything I have not stated.

The final line is the control. Without it, models happily add "all areas were inspected" or "the property was found to be generally sound", neither of which you said. Read the returned text for any sentence that adds a claim, and delete it. A filled-in limitation reads: "We inspected the areas listed in this schedule by eye only. We did not open up fabric or test services. We did not enter the rear shed (locked) or the loft (hatch painted shut), and the under-stairs cupboard was not visible behind stored items." That wording tells a later reader exactly what was and was not recorded.

Choosing the level of detail for the job

Not every schedule needs the same depth. A tenant's end-of-lease baseline, a party recording a neighbour's building before works, and a dilapidations-style record for a commercial unit each need different coverage, and over-detailing wastes time while under-detailing leaves the record thin. Decide at instruction stage and write it into the scope statement:

The ranges are rough guides from the logic of coverage, not standards. Your professional body or your instruction letter may require more. If an instruction says "full" coverage, ask what that means before you start, since the answer sets both your fee and the volume of photos.

Batching large schedules without losing consistency

A big property splits into many prompt sessions, and drift between sessions is a real risk: one session writes "hairline crack", another "fine crack", another "minor crack". Reduce it three ways. First, paste the same style sheet and the same prompt at the top of every session. Second, give a "controlled vocabulary" in the prompt, a short list of allowed terms for common conditions (hairline crack, staining, paint loss, chip, scuff, gap, loose, missing, stiff, worn). Third, after drafting, run a pass that asks the model to list every distinct condition term used in the schedule and to flag near-duplicates. Then replace them with the controlled list.

The same approach applies to reference numbering. Agree the pattern first (floor number, room number, element number) and ask for the AI to use it exactly. Any renumbering after verification risks breaking the link between entry and photo, so freeze the numbers once the surveyor starts initialling.

An example of drift caught late

In a made-up three-storey job, two sessions covered the second and third floors. The second-floor entries used "dated", "tired" and "worn" for carpets; the third-floor ones said "frayed edges" and "thread pulls". During the consistency pass the surveyor found no entry on the second floor mentioned fraying even though the same carpet continued up the stairs. Returning to site, they found frayed edges on the landing and added two entries. The vocabulary pass did not just tidy wording; it surfaced a gap.

When the schedule is used in a dispute

The test of a schedule comes later, when someone points at a mark and says it was there before, or was not. Before issue, ask yourself how each entry would read to a person on the other side. A sentence such as "wall generally in fair condition" gives them nothing to compare. "Eastern wall: no cracks visible, paint intact, one nail hole at 1.4 m height" gives them a baseline they can accept or challenge item by item. The detail is also what makes the record fair to you, because it limits what can later be claimed.

If AI was involved, be ready to say how. You do not need to disclose a tool in the schedule itself, but if asked you should be able to explain that photos were used to draft entries, that every entry was checked and initialled on site, and that the final wording is yours. Keeping the draft, the prompt and the verified version also lets you show what changed between them. Where the schedule is attached to a contract, ask your solicitor whether anything about the use of AI should be stated alongside it.

Slips that weaken an AI-drafted schedule

• Skipping the unphotographed: the AI cannot flag a room it has not seen. Use the room checklist from the template to confirm all elements per room.
• Trusting measurements from photos: a model will estimate size from perspective even without a reference object. The prompt rule on "size not measurable" blocks this; check the output actually follows it.
• Mixing the schedule with advice: clients ask "what should I do about this crack?" The schedule is the wrong place to answer. Offer a separate report.
• Uploading client belongings: a shot of a desk may include names, letters or screens. Crop or avoid them at capture, and keep unedited originals out of AI tools where you can.
• Losing the originals: a schedule built on cropped, renamed copies is hard to defend. Keep the unmodified files in one folder, named to match the entries.

Where the records go next

Once issued, a schedule becomes an attachment, and its use depends on the agreement. Landlord-and-tenant checks work similarly to inventory records, and AI inventory and check-out reports for letting agents shows how a comparison at the end of a tenancy runs against a baseline like this one. Agree with your solicitor how the schedule is referenced in the contract and whether both parties sign it. Store the signed copy, the original photographs and the prompt version together, and keep them as long as your insurer and your agreement require.

The prompt, the template and the verification routine together make the process repeatable. The surveyor remains the author. The AI provides a first draft that is faster to correct than to write, which is a modest gain, and the only one worth having in a document that may be read closely years later.

Questions about AI-drafted schedules of condition

Can AI write a schedule of condition on its own?

It can draft entries from photos, but it cannot see what a photo hides: damp behind a skirting, a loose tile, a smell, a door that sticks. The surveyor has to confirm or correct every entry on site before the schedule is issued, and the surveyor's name is on it, not the tool's.

How many photos does a room need?

Enough that each surface and each defect appears in at least one frame: typically one wide shot per wall, floor and ceiling, plus a close-up and a wider context shot for each defect. A room with a few marks might take 15 to 25 photos.

Should photos with people or belongings go into an AI tool?

Avoid it where you can. Remove or avoid faces, documents and personal items when photographing, and check your tool's privacy settings and your client's permission before uploading anything from an occupied home.

What wording should entries use?

Neutral and measurable: say what is there and where, with sizes where useful, and leave out causes and opinions. "Crack, 120 mm long, hairline, upper left of window head" is a record. "Bad crack from subsidence" is a conclusion.

Is the schedule a legal document?

Often it is attached to a lease or works agreement and may be relied on later in a dispute, so wording matters. Whether it binds anyone depends on the agreement it sits in. Ask your solicitor how it should be referenced in the contract.

Further reads

• Party Wall Notice Template: AI Draft Checked Against the 1996 Act — The notices that often come before a pre-works schedule.
• Use of AI in Building Surveying: Client Data and UK GDPR Basics — Client-data rules when site photos go into AI tools.
• Thermal Imaging Survey Report: Turn Scans Into Plain-English Findings — Another photo-to-report job with the same verify-on-site step.
• RICS AI Standard for Small Firms: Register, Assessment, Client Notice — A professional standard's view on AI use and client notice.
• AI Contract Review for Small Firms: What It Catches and Misses — Check the agreement the schedule is attached to.
• Responsible Use of AI in Surveying Practice: A Small Firm's RICS Checklist — The RICS responsible-AI standard turned into a checklist for a six-person practice, with named owners, filled-in examples and a 90-day plan.
• AI for Building Surveying: A Worked Fee Quote and Enquiry Reply — One enquiry walked end to end: triage, scoping questions, a fee built from hours, the reply, an engagement schedule and the checks before sending.
• Can AI Write an Architect's Snagging List From Site Photos? — A snagging-list workflow for a small practice: photo and note capture, an AI draft by room, severity and responsibility, and a line-by-line check before issue.
• 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.
• Structural Engineer Inspection Report: Photos and Notes to a Draft With AI — A photo-to-report workflow for structural engineers: numbered photos, a register, AI descriptions of what is visible, and checks before the engineer signs.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general surveying and record-keeping practice; ChatGPT and Claude plan pages (privacy settings). No legal advice given.