Yes, AI can turn a technician's drain camera notes into a clear customer report: tidy defect descriptions, photo captions and a plain-words summary. It must not guess severity grades, causes or repair needs. The engineer still checks every distance, defect and caption against the recording, and signs the report before the customer sees it.
The reason this works at all is that most of the effort in a survey report is writing, not judgement. The technician has already seen the pipe, noted the findings and taken stills. What costs time is converting shorthand into sentences, matching captions to pictures, and writing a summary a homeowner or a property manager can follow. That conversion is what an assistant does well, provided it is told not to add anything the notes do not contain.
What goes in, what comes out, and who checks
A reliable flow has four parts, and the order matters.
- The technician records notes during the survey in a fixed shape: distance, clock position where relevant, defect, remark, photo number.
- The notes and a short job header go into the assistant with a prompt that forbids guessing.
- The assistant returns a draft customer summary, a defect table and photo captions, with every gap marked.
- The engineer checks the draft against the recording and stills, fills the gaps, applies any grades under the coding scheme in use, and signs.
Steps 1 and 4 are human. Steps 2 and 3 take minutes. If you skip step 4 you have an unsupervised author, and in a survey report an invented sentence can end up in a dispute between a customer and a neighbour or an insurer. A sibling tutorial on sewer camera reports from defect list to customer summary covers the defect-list side; this one concentrates on notes-to-report and the sign-off checks.
A worked example: from shorthand to a draft report
An illustration. A small drainage firm surveys a rear drain run of about 14 m from an inspection chamber to a public sewer connection at a terraced house. The technician's notes, as typed into a phone on site, read:
Job: rear drain, chamber 1 to sewer connection, 100mm clay, approx 14m.
0.0 start at chamber 1, flow ok, light scale
2.1 jt displaced 3 o'clock, small, photo 3
4.8 crack long. 12 o'clock, ~300mm, photo 5
7.4 root ingress jt 10 o'clock, fine roots, photo 7
7.9 water pooling, sag? (check vid)
11.2 jt open slight, photo 9
13.6 lateral from left, 100mm, capped?
14.0 end at sewer conn, fully visible, no blockageThose notes are fast to write and slow to read. They also contain uncertain items (a possible sag, a lateral that might be capped) that a careful report must not turn into facts.
A prompt that forbids grades and causes
You are drafting a customer drain survey report for a plumbing and drainage firm.
Use ONLY the technician's notes below. Do not add any defect, cause, grade, cost or recommendation that is not in the notes.
Output three parts:
A. A defect table: distance, position, observation in plain words, photo number, status (CONFIRMED or ENGINEER TO CHECK).
B. A one-paragraph customer summary, in plain words, no jargon, no alarmist wording.
C. A list of items the engineer must confirm before sending, including anything marked with a question mark in the notes.
Where a grade or severity would normally go, write "grade to be assigned by engineer".
Notes: [paste notes]
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowDraft defect table and summary (illustrative)
Part A, as a table.
| Distance | Position | Observation | Photo | Status |
|---|---|---|---|---|
| 0.0 m | Chamber 1 | Start of survey. Flow normal. Light scale on pipe wall | None | Confirmed |
| 2.1 m | 3 o'clock | Joint displaced slightly | 3 | Confirmed |
| 4.8 m | 12 o'clock | Longitudinal crack, about 300 mm long | 5 | Confirmed |
| 7.4 m | 10 o'clock | Fine roots entering at a joint | 7 | Confirmed |
| 7.9 m | Invert | Water standing in the pipe; possible dip in the pipe, not yet confirmed | None | Engineer to check |
| 11.2 m | Joint | Joint slightly open | 9 | Confirmed |
| 13.6 m | Left side | Side connection, 100 mm; whether it is capped is not stated | None | Engineer to check |
| 14.0 m | Sewer connection | End of survey. Connection fully visible, no blockage | None | Confirmed |
Part B, the customer paragraph: "We surveyed the rear drain from the first chamber to the connection with the main sewer, about 14 metres of 100 mm clay pipe, using a camera. The pipe is flowing and the connection at the far end is clear. We recorded several joint and pipe defects along the way: one displaced joint, a crack about 30 cm long, fine roots entering at one joint and a slightly open joint. We also saw standing water at about 8 metres, which we are checking to see whether the pipe has dipped. Grades for each defect will be added by our engineer."
Part C: "Confirm the standing water at 7.9 m (sag or normal pooling) from the recording. Confirm whether the side connection at 13.6 m is capped. Assign grades. Check the photo numbers match the stills. Confirm pipe material and diameter from the recording."
Two corrections made from the recording
Two edits are typical. First, the draft says "about 14 metres" while the notes said "approx 14m" and the end point is 14.0 m; the engineer confirms the measured length on the equipment display and changes the wording to the exact figure. Second, the customer paragraph says "fine roots entering at one joint" but the video shows the roots also at 7.6 m, a distance the technician missed in the notes. Notes-to-report cannot find what the notes omit, which is why the engineer scrubs the recording, not just the draft. That is the real sign-off: the draft is checked against the evidence, and the evidence is the video.
Defect wording that stays inside the notes
Careful wording is half the value. Three rules keep the report honest.
Describe, don't diagnose. "Fine roots entering at a joint" is an observation. "The drain is failing because of the neighbour's tree" is a diagnosis. The first is supported by a still; the second needs reasoning the engineer has to own.
Use calibrated verbs. Seen, recorded, noted, appears, possible. If the technician wrote a question mark, the report keeps a question mark, in words: "possible dip, to be confirmed". A model that is told to keep uncertainty visible will do so; one that is not told will smooth it over.
Keep severity and cost out of the draft. Condition grades come from a defined coding scheme applied by a trained person to the footage. Repair cost is a quote, not a survey finding. Both belong to later steps. Our tutorial on using AI to write plumbing quotes picks up from the point where findings become priced options.
Photo captions that match the pictures
Captions are a quiet source of errors. The technician's photo numbers in the notes need to match the file numbers in the report, and the caption needs to say what the picture actually shows. Here is a before and after, illustrative.
Before, from the notes: "photo 5 crack long 12 o'clock 300mm". After, as a caption: "Photo 5: longitudinal crack at the top of the pipe, about 4.8 m from the start, roughly 300 mm long." The after version is longer, but it works for a reader who has not seen the video. If you can upload the stills, ask the assistant to compare each caption with its picture and flag mismatches, such as a caption that says "crack" over a photo of a joint. Treat that as a second look, not as proof, and open the stills yourself for anything flagged.
Survey jobs that change the drafting routine
A technician who writes in full sentences
Some technicians write longer notes. The same prompt works, and the output needs fewer edits. The risk changes: long notes may contain opinions ("clearly the neighbour's work"). Add a rule: "If a note contains an opinion about cause or blame, remove it and list it under items for the engineer."
A survey that could not be completed
If the camera hit a blockage at 9 m, the notes may say "stopped, blocked, can't pass". The draft must say the survey was incomplete and why, and must not describe the rest of the run. Add: "If the survey did not reach the end, say so in the first line of the summary."
Repeat customers and portfolios
A property manager with twenty units wants a consistent format. Keep the prompt in a saved project or snippet so every report uses the same table, the same plain-words rules and the same engineer list. Claude Projects and ChatGPT Projects both hold standing instructions and files that every report can reuse; ChatGPT Business is $25 a seat monthly or $20 on annual billing, with a two-seat minimum shared with Claude Team.
Reports that may be used in a dispute
If a report may be used by an insurer, lawyer or a neighbour, tighten the process: keep the technician's original notes, the AI draft, the engineer's marked-up version and the recording together, and state in the report who inspected and who checked. Accuracy matters more than polish here, and the engineer's name on the report is the point. Ask a qualified adviser if you are unsure about your duties.
Confidential property details
Addresses and customer names do not need to go into a chat to produce a draft. Replace them with "[address]" and "[customer]" in the prompt, then add them back in your document. On business plans, ChatGPT and Claude do not train on your content by default, but removing names costs a few seconds and removes the question.
A second example: a survey that stopped short
Not every survey reaches the end, and incomplete runs are where careless drafts do the most harm. Here are a technician's notes for a different job, illustrative.
Job: front drain, chamber 2 to boundary, 150mm concrete, planned 20m.
0.0 start, heavy scale
5.5 scale heavy, narrowing?
9.2 obstruction, hard, camera cannot pass, photo 4
Stopped. Customer told on site, jetting quoted separately.A careless draft would write a smooth summary of "a 20 m drain with scale and an obstruction". The prompt rule from earlier, "if the survey did not reach the end, say so in the first line", produces something closer to this: "The survey was not completed. The camera stopped at 9.2 m because of a hard obstruction and could not pass, so about 10.8 m of the planned 20 m was not inspected. Heavy scale was seen from the start of the run." The table has three rows, and Part C reminds the engineer to state that the condition beyond 9.2 m is unknown, to check whether "narrowing?" at 5.5 m is visible on the recording, and to confirm what the customer was told about jetting.
Note what the draft did not do. It did not say the pipe was blocked with a cause, and it did not offer a view on whether the rest of the run is sound. Those absences are the behaviour you are paying the prompt for.
A technician's notes template that makes drafting easier
The better the notes, the less the assistant has to interpret. A one-line template costs the technician almost nothing and removes most ambiguity.
[distance m] | [clock position or invert/crown] | [defect word] | [size or extent] | [photo no.] | [sure / not sure]Filled in, illustrative: "4.8 | 12 | crack longitudinal | 300 mm | 5 | sure". The final field matters most. Anything marked "not sure" goes straight into the engineer's list, without the assistant needing to guess which notes were uncertain. A team that adopts this template usually finds the question marks in their notes stop appearing in odd places, because there is now a proper spot for doubt.
How a question mark became a capped connection
A realistic slip: the technician writes "13.6 lateral from left, 100mm, capped?" and the draft, tidied too enthusiastically, reads "a capped side connection at 13.6 m". The question mark vanished and a doubt became a fact. If that report goes out and the connection turns out to be live, the customer has been told something untrue about their own drains. This is why Part C of the prompt exists, and why the engineer's check lists every question mark. Search the draft for the words capped, blocked, collapsed, failed and live; each one is a claim that needs a recording to back it.
Using stills and the recording honestly
A general chat assistant works from what you give it. If you upload five stills, it describes five stills; it has not watched the video. Say so in the report process rather than letting anyone assume the tool "saw" the drain. Specialist software exists that analyses survey footage itself, and a firm that adopts one should apply the same discipline: its output is a proposed coding that an engineer confirms. Whichever you use, the line in the report that matters is who looked at the recording and when.
Image limits also apply. If a still is dark or the camera lens was wet, the assistant may describe detail that is not clearly visible. Add a prompt line, "if a photo is unclear, say it is unclear and describe nothing", and compare any caption the assistant writes with the still before it goes in the report.
One more safeguard is worth the thirty seconds it takes: ask the assistant to read its own draft back against the notes. "List every statement in the draft that is not directly supported by a line in the notes." A good model will point at the soft spots, such as the phrase "light scale" being softened to "minor build-up", and you can reword them. It is a second pass for catching drift, and it does not replace the engineer, because the model is checking its draft against your notes, not against the pipe.
The engineer's sign-off checks
Print this and use it for every report that leaves the firm.
| Check | How | Done |
|---|---|---|
| Header matches the job | Address, date, pipe material and diameter against the work order and recording | Initial |
| Every distance matches | Defect table against the equipment's distance counter or the recording | Initial |
| No defect missing | Scrub the recording at normal speed, not only to the noted times | Initial |
| No defect added | Each table row traced to a note or the recording | Initial |
| Uncertain items resolved | Each "engineer to check" row confirmed or reworded | Initial |
| Grades applied by the engineer | Using the coding scheme the client or authority requires | Initial |
| Captions match stills | Open each photo beside its caption | Initial |
| Summary has no cause, blame or cost | Read the customer paragraph for diagnosis words | Initial |
| Limits stated | Incomplete runs, blocked sections and what the camera could not see | Initial |
| Signed and filed | Name, date, files saved with the original notes | Initial |
One completed row, as an illustration: "No defect missing: recording scrubbed at normal speed, a second root ingress at 7.6 m added to the table, initialled by the engineer on the day." A report with ten initialled lines is a different object from a document that merely looks professional.
Where the time goes and whether it pays
Say a firm does eight surveys a week. Writing each report by hand takes about 35 minutes, so 4 hours 40 minutes. With drafting, the notes still take their usual time, the draft takes five minutes, and the engineer's check takes 15 to 20 minutes. That is roughly 25 minutes per report, saving about 10 minutes each and 80 minutes a week. The larger gain is consistency: every report has the same table, the same uncertainty flags and the same plain-words summary. The risk is that checks get shortened when the draft looks good. A draft that looks finished is more dangerous than one that looks rough, so keep the checklist in the process and not in anyone's memory.
Where this fits among other jobs
Notes-to-report is one of the safest early uses of AI in a drainage or plumbing firm, because the source material is yours and the output is checked line by line. The same shape suits other test paperwork: our tutorial on backflow test reminders and reports covers a case where the form is fixed and the technician supplies the numbers. When a survey leads to extra work, the change-order examples help you describe it without disputes, and the wider 30-day plan for a plumbing business shows where to put report drafting in the order of adoption.
If you start small, run five old surveys through the flow, compare the drafts with the reports you actually sent, and note every sentence the assistant invented. That list becomes your prompt rules. Once the invented sentences stop appearing in the drafts, the checks can stay exactly as they are, because they are what keeps it that way.
Questions about AI and drain survey reports
Can AI watch the drain camera video and write the report?
Some specialist software analyses footage, but a general chat assistant works from your notes and photos, not from the video. For most small firms the safest route is the technician's notes plus labelled stills, turned into a draft that the engineer checks against the recording before it goes out.
Should AI assign severity or condition grades to defects?
No. Grades come from a defined coding scheme applied by a trained person looking at the footage. A model will happily invent a plausible grade. Leave the grade fields for the engineer, and have the draft say 'grade to be confirmed' until then.
Can a customer report mention the likely cause of a defect?
Only if the engineer has judged it from the evidence. Notes saying 'root ingress at joint' support a description of what was seen. A statement of cause or of what the repair must be goes beyond the notes and should come from the engineer, in their own words.
How long should a customer summary be?
One page for the customer is plenty: what was inspected, what was found, what it means in plain words, and the options. Keep the full defect table and photos as an appendix for the engineer, insurer or solicitor who may need them.
Further reads
- 10 ChatGPT Prompts for Plumbers: Call-Backs, Quotes, Complaints — Ten ChatGPT prompts for callbacks, quotes and complaints.
- Plumbing Maintenance Plans: AI-Written Offers and Renewals — Offer follow-up drain checks as a maintenance plan.
- AI Quoting Tools for Plumbers: What They Cost and What They Do — What AI quoting tools cost and do for plumbers.
- AI Voice Agent for Plumbers: A 15-Call Test Before It Takes Real Jobs — Test an AI that takes the call before the survey is booked.
- How Surveyors Use AI to Turn Site Notes Into Reports — Another trade turning site notes into reports.
- Can AI Tools Help With Plumbing Design? What Needs a Qualified Person — Where a qualified person must sign off AI drafts.
- AI Marketing for Plumbers: A Worked 12-Month Posts and Offers Calendar — A worked 12-month calendar for a five-van plumbing firm: frozen pipes, boiler checks, heater age and drain season, with AI drafts and a review step.
- Damp Survey Report: Write Findings and Quote Lines With AI — Turn readings and site notes into a clear damp report and quote lines in ChatGPT or Claude, while the surveyor keeps the diagnosis and checks every line.
- 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.
- 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.
- Chimney Inspection Report: Can AI Turn Camera Findings Into Plain English? — How a chimney sweep can turn camera survey notes into a plain-English customer report with AI, and what must still be recorded by hand.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: vendor pricing, privacy and file-upload pages for ChatGPT and Claude (checked 1 Nov 2026); general drain survey practice. No coding standard is quoted: use the scheme your clients and the relevant authority require.