AI Can Write a Convincing Survey Report. That’s Exactly the Problem.

There is something slightly uncomfortable about modern AI: it can be wrong very convincingly. Give a generative AI system a handful of inspection notes and ask it to turn them into professional survey commentary, and within seconds it can produce something remarkably polished. The grammar is good, the terminology sounds appropriate and the paragraph may read exactly as you would expect professional advice to read.

For residential surveyors, however, that creates an important problem. A paragraph that sounds professional is not necessarily an accurate reflection of what was inspected, observed or concluded. The more convincing the language sounds, the easier it can be to miss when AI has introduced information that was not supported by the evidence.

The obvious mistakes are the easy ones

If an AI system produces something obviously ridiculous, an experienced surveyor is unlikely to be fooled by it. The real concern is what happens when the mistake isn’t obvious at all, but sounds entirely reasonable.

Take a simple inspection note: “Hairline cracking noted above the rear bedroom window.” An AI system could plausibly expand this into: “The cracking above the rear bedroom window is likely to be associated with slight structural movement around the opening.”

It sounds entirely reasonable. But the surveyor recorded the presence and location of cracking; they did not establish that structural movement was responsible. In a few polished words, the AI has turned an observation into a diagnosis.

This distinction matters because a small change in wording can significantly alter what is being communicated to the client.

When plausible becomes fact

Generative AI is designed to produce coherent language. When information is incomplete, it may fill the gap with something plausible. In many everyday applications that can be useful. In professional reporting, it requires much tighter boundaries.

The same problem can arise when AI drafts location commentary. It might state that a property is “situated within a designated conservation area” when this, in fact, is not the case. This is a form of AI hallucination, where plausible information is presented as fact.

Surveying is full of these distinctions. An older electrical installation may be observed and recorded during an inspection, but its age or appearance alone does not establish that it is unsafe or requires replacement. Reaching such a conclusion requires appropriate evidence and, where necessary, further investigation by a suitably qualified person.

Professional reporting is therefore about much more than converting observations into well-written sentences. It requires understanding what the available evidence supports, what remains uncertain and where further investigation may be appropriate.

Perhaps we’re asking the wrong question

Much of the discussion around AI in surveying starts with: “Can AI write a survey report?”

Technically, it can produce text that resembles one. But perhaps the more important question is: what should AI be allowed to add that wasn’t contained within the information provided to it?

There are many areas where AI can provide genuine assistance. Transcribing inspection recordings, organising information, improving grammar, retrieving relevant material and helping structure draft commentary are very different from independently diagnosing defects or introducing assumptions about a property.

The objective shouldn’t be to avoid AI. It should be to use it with appropriate boundaries.

Five checks before AI-generated text reaches your report

If you use AI within your reporting process, here are five simple checks worth applying before generated commentary reaches the final report.

1. Ask where the information came from. Was it recorded during the inspection, obtained from a reliable source or based on an opinion you have actually formed? If you cannot identify the source, take another look.

2. Watch for observation becoming diagnosis. “Cracking was noted” and “the cracking is caused by structural movement” are very different statements. Check whether AI has introduced a cause or conclusion that wasn’t established.

3. Verify external facts. Conservation-area status, flood risk, planning information and other property-specific facts shouldn’t be accepted simply because AI states them confidently. Check material information against an appropriate source.

4. Check the reasoning behind the recommendation. An unsupported assumption can lead to inappropriate advice. If AI incorrectly assumes the cause of staining, for example, the recommendation that follows may also be wrong.

5. Review accuracy before wording. Don’t let polished language distract from the substance. Before adjusting grammar or style, ask whether the commentary is supported, proportionate and consistent with what you actually found.

The principle is simple: AI-generated commentary should be treated as a draft to review, not evidence to rely upon.

Good reporting is still about professional judgement

The value of a survey report doesn’t come from elegant prose. It comes from the professional decisions behind it: what matters, what the evidence supports, what the client needs to understand and whether further investigation is appropriate.

AI can help organise and communicate those decisions. What it shouldn’t do is manufacture them.

And this is where increasingly capable AI creates an interesting challenge. The better the writing becomes, the less obvious an unsupported statement may appear. The most problematic hallucination may not be something absurd. It may be something perfectly plausible that simply isn’t true.

AI should work around the surveyor, not the other way around

The risks discussed above are exactly why we have taken a deliberately controlled approach to AI within MyReport®. Report Genius® works from the surveyor’s own inspection information and reporting content, with the surveyor remaining responsible for reviewing and approving what reaches the report. AI can be used selectively where it adds value, rather than being allowed to take over the reporting process.

And importantly, AI is optional. Turn it off and MyReport® remains a complete reporting workspace, with voice capture, site notes, photographs and mark-ups, your own reporting content and CoReview™ for collaborative review. Where AI is used, data remains private and is not used to train public AI models.

The principle is simple: use AI to support the reporting process, not to replace the expertise behind the report.