Export the surveyor's defect list as rows (distance, code, grade, remarks), give AI only those rows with fixed rules, and have it draft a plain-English summary grouped by urgency. The surveyor then checks every figure against the list and the footage before signing. AI explains the coding; it never codes, grades or judges footage.
When AI drafts a sewer camera inspection report, the usual failure is softening and adding, not typos: a grade 4 fracture becomes "a minor crack", or a root intrusion acquires a cause, such as "from the oak tree", that the surveyor never wrote. A report a customer relies on for a purchase, a repair quote or an insurance claim can carry neither. The method below is built around rules that block both and a checklist that catches what slips through.
Start from the defect list, cleaned and anonymised
Everything downstream depends on one table: the observations the surveyor coded, exactly as the survey software exports them. For each run you need the run details (start and finish chambers, length, pipe material and diameter, direction of survey) and one row per observation with distance, clock position, code, grade, the surveyor's own remarks and, if your standard uses it, a band the surveyor assigns from the grade. Keep the video timestamp in your file for checking, but the assistant does not need it. Strip names, addresses and phone numbers and use a job reference instead.
Here is an illustrative run, 38 m of 150 mm clay pipe surveyed with the flow from the rear access chamber to the street chamber. Assume the surveyor maps grades 4 and 5 to band A, grade 3 to band B and grades 1 and 2 to band C. The mapping belongs to the surveyor. The assistant applies it and never changes it.
| # | Distance (m) | Clock | Observation | Grade | Band | Surveyor's remarks |
|---|---|---|---|---|---|---|
| 1 | 6.4 | 12 | Joint displaced, medium | 2 | C | Offset about 10 mm, no soil visible |
| 2 | 11.8 | 10 | Roots at joint, fine | 3 | B | Fine roots entering, about 10% of cross-section |
| 3 | 14.0 | 3 | Crack, longitudinal | 2 | C | Hairline, 0.4 m long |
| 4 | 19.5 | 6 | Deposits, settled | 2 | C | Light silt, under 5% |
| 5 | 22.3 to 24.1 | 6 | Water level, standing | 3 | B | About 20% depth over 1.8 m, possible sag |
| 6 | 27.9 | 9 to 3 | Fracture, circumferential | 4 | A | Wall broken all round, soil visible at crown |
| 7 | 31.2 | 12 | Roots, mass | 4 | A | About 40% of cross-section blocked |
| 8 | 34.0 | 3 | Connection, intruding | 2 | C | Branch pipe intrudes about 15 mm |
Alongside the rows, supply the limits of the survey in the surveyor's words: "Run surveyed with the flow only. No dye test. Connections not traced individually. Finish chamber reached at 38.0 m." And supply the surveyor's recommendations as a separate field, even if it is short. If that field is empty, the summary should say that recommendations are still to be added by the surveyor, not invent some.
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 nowA comment library the surveyor owns
The safest way to get plain English is to write it once, carefully, and reuse it. Ask the assistant to draft a first version of a library, one approved sentence per observation type, then have the surveyor edit every line. After that, the library goes into the prompt, and the assistant's job shrinks to arranging approved sentences around the numbers. Building a comment library for home inspection reports with AI covers the same technique for another survey, and it carries over directly.
| Observation | Approved plain-English sentence |
|---|---|
| Joint displaced | Two pipe sections have slipped out of line at a joint. |
| Roots at joint | Plant roots have grown into the pipe at a joint. |
| Crack, longitudinal | A crack runs along the length of the pipe. |
| Fracture, circumferential | The pipe wall is broken all the way round. |
| Deposits, settled | Silt or debris has settled in the bottom of the pipe. |
| Water level, standing | Water is sitting in the pipe instead of draining away. The surveyor's remarks say whether a dip or debris is suspected. |
| Connection, intruding | A branch pipe pushes into the main pipe. |
Look at what the library avoids. "Plant roots" instead of "tree roots", because the footage shows roots, not which tree they belong to. No sentence says why a defect happened. And the water-level sentence points back to the surveyor's remarks, so the assistant cannot upgrade "possible sag" into a sag. Twenty to thirty entries cover most surveys, and the library only needs revisiting when your coding standard or your wording changes.
The prompt: rules that stop the assistant adding anything
Save this in a Project in ChatGPT or Claude, or a Gemini Gem (becoming a skill from November 2026), so the rules are fixed and only the run data changes. The library and the surveyor's band mapping sit in the saved instructions.
You turn a surveyor's defect list into a plain-English summary for a customer. A named surveyor will check and sign it.
Rules:
1. Use only the rows, limits and recommendations below. If it is not in the data, it did not happen.
2. Copy every distance, grade, percentage and length exactly as given. Do not round.
3. Group findings by the Band column (A serious, B moderate, C minor). Never move an item between bands. Never call an item minor unless its band is C.
4. Describe each observation with the approved sentence for its type from the library, then add its distance, grade and the surveyor's remark.
5. Do not state causes, repair methods, costs, timescales or legal consequences unless they appear in the remarks or recommendations.
6. Do not write "safe", "fine", "no problem" or "guarantee".
7. State the limits of the survey in their own section.
8. Under 300 words, sentences under 25 words, no jargon beyond the library.
9. End with a list headed FOR THE SURVEYOR TO CONFIRM naming anything unclear, ambiguous or missing.
Run details: [paste]
Observations: [paste table]
Limits: [paste]
Surveyor's recommendations: [paste, or write NONE]
Approved library: [paste]Each rule answers a known failure. Rule 2 blocks rounding, rule 3 blocks softening, rule 5 blocks invented causes and costs, and rule 9 gives the assistant a safe place for doubt, so it flags an unclear item instead of resolving it with a guess. Write the rules in the saved instructions once and leave them alone. When a new failure appears, add a rule for it rather than rewording the old ones.
What the first draft looked like, and what the surveyor changed
Even with those rules, a first draft needs reading line by line. This one, from a plausible run on the table above, is illustrative, and it contains the mistakes that turn up most often:
Summary for job [ref]. The camera reached the end of the run and recorded eight observations.
Serious: at 27.9 m the pipe wall is broken all the way round and soil is visible at the top of the pipe (grade 4). At 31.2 m roots block about 40% of the pipe (grade 4). Repairs of this kind commonly run into the thousands.
Moderate: at about 12 m, fine roots from nearby trees are growing into a joint. At 22.3 to 24.1 m the pipe has sagged, so water sits in it. Both are fairly minor.
Minor: a joint slightly out of line at 6.4 m, a hairline crack at 14.0 m, light silt at 19.5 m and a branch pipe pushing into the main pipe at 34.0 m.
The drain is otherwise working normally. Limits: run with the flow only; no dye test.
- "About 12 m" becomes 11.8 m. Rounding breaks rule 2, and a repair crew digging from the report needs the exact figure.
- "From nearby trees" comes out. The footage shows roots, and it cannot show which plant they belong to.
- "Has sagged" becomes "possible sag", the surveyor's own words. The assistant hardened a doubt into a fact.
- "Fairly minor" comes out. Both items are band B, which is moderate, and the customer must not read them as ignorable.
- "Repairs of this kind commonly run into the thousands" comes out. It is a cost the surveyor never wrote, and a customer may quote it back.
- "Otherwise working normally" comes out. Nothing in the data says anything about flow capacity beyond the listed observations.
- The limits are incomplete. "Connections not traced individually" is missing and goes back in.
Notice that the count of eight observations is correct, and that the two serious items are in the right band. The assistant got the structure right and the wording wrong in the places where wording matters, which is why the rules target wording and the checks target numbers. The same discipline of severity order appears in a snagging report template with sections, severity order and AI captions.
A layout the customer can read in two minutes
Most customers read the first paragraph and the next steps, then forward the rest to a builder, a solicitor or a buyer. Build the layout around those two reads. Five sections, always in the same order, make the report easy to scan and easy to check:
- Verdict. One or two sentences: how many observations, how many in each band, and which items need repair.
- What we found. Three bands, most serious first, each item with its distance, grade and approved sentence.
- What the surveyor recommends. Taken only from the recommendations field.
- What we could not see. The limits of the survey in plain words.
- Attachments. The full defect log, a sketch if you produce one, and the footage link.
Filled in for the run above, the first and third sections read as follows (illustrative):
Verdict: We recorded eight observations in the 38 m run, measured from the rear chamber towards the street: two serious, two moderate and four minor. The two serious items, at 27.9 m and 31.2 m, need repair.
The surveyor recommends: obtain repair quotes for the fracture at 27.9 m and the roots at 31.2 m; re-survey the section from 22.3 to 24.1 m after repairs to see whether the standing water remains; monitor the minor items at the next survey.
One line in that verdict earns its place: "measured from the rear chamber towards the street". State which chamber the distances run from, and in which direction the camera travelled. A repair crew that assumes distances run from the street end would measure 27.9 m from the wrong chamber and dig about 18 m away from the fracture, and the report would have done nothing to stop it. Ask the assistant to define each term in a clause the first time it appears and to keep sentences under 25 words. Then read the verdict aloud. If you would not say it that way to the customer on the phone, rewrite it.
Checks the surveyor runs before signing
The checks take about ten minutes and are the reason the method is safe. They are mostly arithmetic and comparison, so do them in order, and tick each one.
- Count the observations in the summary and compare the number with the rows in the list. Same count, same bands.
- Check every distance, grade, percentage and length character by character against the list.
- Look for banding words. "Serious" appears only for band A, "minor" only for band C, and "moderate" for band B.
- Look for anything the data does not hold: a cause, a repair method, a cost, a timescale, a legal consequence. If it is not in the remarks or recommendations, cut it.
- Check the limits section. Is everything the camera could not do or see stated, including aborted lengths and untraced connections?
- Re-watch the footage at each band A item and at one other item, and confirm that the sentence describes what you see.
- Check the direction of survey, the chamber names and the units against the run details.
- Read the verdict as the customer would. Could a reasonable person act on it wrongly?
Search the finished text for a short list of red-flag words: safe, fine, no problem, guarantee, caused by, will collapse, minor (outside band C). Each hit needs a reason or a rewrite. Can ChatGPT write home inspection report narratives, and what to check goes through the liability side of the same issue, and it is worth reading before you sign the first AI-assisted report.
Different readers, different summaries
The defect list does not change, but the person reading the report does. Keep one saved prompt and add a single instruction line per audience, so the facts stay identical across versions.
| Reader | What they need | The extra prompt line | The trap |
|---|---|---|---|
| Home buyer | Findings in order of seriousness, with distances a contractor can find | "Put band A and B first and give the distance for every item." | The tone drifting into advice on price or negotiation |
| Landlord or property manager | A works list and a re-inspection point | "Add a table of items with distance, band and the surveyor's recommendation." | The assistant inventing deadlines such as "repair within 30 days" |
| Insurer or loss adjuster | Facts only, with no opinion | "Remove recommendations. List observations and limits in neutral wording." | Any sentence that implies what caused a loss |
| Commercial manager with several runs | A comparison across runs | "Open with a table of bands per run, then one paragraph per run, with the run ID on every line." | Mixing runs, since distances repeat from run to run |
Notice that none of the extra lines asks the assistant to know more. Each one only changes the arrangement or removes something. Every audience-specific addition of facts, such as a recommendation to re-line the pipe or a view on how a defect arose, has to come from the surveyor, not from the prompt.
Surveys that stop early or find nothing
The two awkward reports are the ones where the camera could not finish and the ones where nothing was found. In both, the dangerous sentence is the reassuring one. Add a rule to the prompt that says: if the run was not completed, the first sentence of the summary says so and the verdict gives the length not inspected. Three illustrative wordings the surveyor can keep in the library:
Stopped early: The camera could not pass 18.6 m because of [obstruction recorded by the surveyor]. The remaining 19.4 m of the 38.0 m run was not inspected, so this report says nothing about its condition.
No defects recorded: No defects were recorded in the 24.0 m surveyed. This describes the pipe on the day of the survey, and only the length the camera reached.
Part full pipe: Between 10.0 m and 12.5 m the pipe was part full of water, so the lower part of the pipe could not be seen. Any defect below the water line was not visible.
A report of the second kind needs special care, because customers read "no defects" as "no problem". The sentence about the day of the survey and the length reached is what separates a finding from a guarantee. If your client contract or insurer expects particular limitation wording, ask a solicitor to approve it once and put it in the library, so the assistant never writes its own.
Comparing a re-survey with the original after repairs
A second survey after repair work produces the report customers look at hardest, because they paid for the fix. The new list rarely matches the old one exactly. Distances shift by a few centimetres, a repair adds its own observations, and a defect that simply is not mentioned may or may not have been repaired. Give the assistant both lists and one extra rule: absence from the new list means "not recorded in the new survey", never "repaired". Only the surveyor can say a defect was repaired, and only if the new list or the contractor's record shows it.
| Item in the first survey | Second survey | Safe wording for the summary |
|---|---|---|
| Fracture at 27.9 m (band A) | Patch repair recorded from 27.8 to 28.2 m | The fracture at 27.9 m is not recorded in the second survey. A patch repair is recorded from 27.8 to 28.2 m. |
| Roots at 31.2 m (band A) | Nothing recorded | The roots at 31.2 m are not recorded in the second survey. For the surveyor to confirm: whether the roots were cut or the pipe was repaired. |
| Standing water, 22.3 to 24.1 m (band B) | Still recorded, 22.3 to 24.0 m, about 20% depth | Standing water is still recorded from 22.3 to 24.0 m, at about 20% depth. |
The middle row is where assistants go wrong. A tidy draft says "the roots have been removed", which is a statement about work done that no row supports. The wording above reports what the camera saw, and routes the open question to the surveyor, who can ask the contractor. If a lining or patch has been installed, the surveyor's remarks should say so, and the library needs a sentence for it that the surveyor has approved.
Minutes saved per report, and what stays out of the chatbot
A sum, with illustrative minutes. Writing a customer summary by hand takes about 25 minutes. With the library and the saved prompt, preparing the data takes about two minutes, the run about one and the checks about ten, so 13 minutes in all. That saves 12 minutes a report, and for a firm that issues six reports a week it is 72 minutes. If your own saving looks much bigger than that, check that you have not dropped a check. The checks are the product, and the typing is the cheap part.
Privacy needs a firm rule. Names, addresses, phone numbers and footage links stay out of the assistant. Use a job reference, and add the customer's details to the finished document inside your own report template. The footage and the coding stay in the survey software, so the assistant only ever receives rows of text. On personal accounts, training use of your chats can be switched off in the privacy settings, while business plans of ChatGPT, Claude and Gemini in Workspace say they do not train on business content by default. Setting labels change, so look at yours. AI security risks for small businesses and how to close them lists the other settings to check.
If you have no coded list and are starting from rough notes, turning drain camera notes into customer reports with AI covers that case, and chimney inspection reports that turn camera findings into plain English shows how a neighbouring trade handles the same step. Whichever route you take, the rule from the start of this tutorial holds: footage and defect coding stay the surveyor's, and a named person signs every report.
Questions drain survey firms ask about AI summaries
Can AI code the defects from the footage?
Coding stays with the surveyor. It is the professional judgement the customer is paying for, and the part that can be challenged later. If a tool you use offers to suggest codes, treat its output as a draft that the surveyor reviews, changes and signs for. The method here starts after the coding is finished, with the surveyor's list as the only input.
Should the report say that AI helped write the summary?
Follow your client contract, your insurer's requirements and any professional body guidance, and ask a solicitor if you are unsure. Whatever the policy, a named surveyor signs the report and has checked every line against the defect list and the footage. That check, not the tool, is what the customer is relying on.
Do I need to paste addresses and names into the assistant?
No. The summary needs only the observation rows, the run details, the surveyor's recommendations and the limits of the survey. Use a job reference in place of names, addresses and phone numbers, and add the customer's details to the finished document afterwards, in your own report template.
How long should the customer summary be?
Aim for one page: a one-sentence verdict, the findings in three urgency bands, what could not be seen, and the surveyor's recommended next steps. The full defect log and the footage links go in the attachments. A reader who stops after the first paragraph should still understand how serious the findings are.
Further reads
- Damp Survey Report: Write Findings and Quote Lines With AI — Another survey where findings become a report and quote lines.
- Thermal Imaging Survey Report: Turn Scans Into Plain-English Findings — Plain-English findings from scans, with the same checks.
- Roof Inspection Reports: Photos to a Customer-Ready Report With AI — Photos to a customer-ready report in a neighbouring trade.
- Septic Inspection Report: Can AI Write It From Field Notes? — Field notes to an inspection report for tank systems.
- Mould Inspection Report Template: Findings, Photos and Limits With AI — Findings, photos and stated limits in a mould report.
- Structural Engineer Inspection Report: Photos and Notes to a Draft With AI — How an engineer drafts a report from photos and notes.
- 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 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.
- Can AI Tools Help With Plumbing Design? What Needs a Qualified Person — What AI can draft in a plumbing layout, what a licensed designer must verify, and a sign-off checklist for the contractor.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: the privacy and data-use pages of ChatGPT, Claude and Gemini. The defect list, comment library, report wording and time figures are illustrative, and your own coding standard and client contract decide what a report must contain.