Key takeaways
- AI in surveying software drafts text and selects clauses; it does not exercise professional judgement or assign condition ratings
- The surveyor's liability for every word in a signed report is unchanged by how much AI assisted in writing it
- A practical split exists between what you can reasonably delegate to AI assistance and what you must personally review before sign-off
- Before relying on any software, check what it logs, what it flags, and how easy it makes the review step
What AI decisions in a RICS Home Survey actually look like
When people talk about AI decisions in RICS Home Survey report production, they usually mean one of two different things: the software making a professional judgement call, or the software drafting text based on the surveyor's own observations. These are not the same thing, and conflating them causes real confusion about where the risk sits.
Current AI in RICS home survey software does the second of those two things. It reads your site notes, matches them to known patterns, pulls relevant standard clauses from a library, stitches them into report language, and flags where your notes seem incomplete for a given element. That is genuinely useful. It can cut drafting time substantially and reduce the chance of a section being left blank.
What it does not do is decide whether a crack is structural or cosmetic, whether a flat roof has three or seven years of remaining life, or whether a condition rating of 2 or 3 is appropriate for a particular defect. Those calls require the kind of contextual, experience-based judgement that current AI cannot replicate. A language model has no eyes, no nose, no feel for the way a floor bounces underfoot. It has text.
There was a Level 3 on a 1930s semi last winter where the AI draft came back with a perfectly competent paragraph on the chimney stack, appropriate language, the right advice about repointing. What it could not know was that I had stood in the loft and felt the cold air coming through at the base of that stack, and that the flashing visible from the rear elevation was not original lead but a poorly fitted cover that had been painted over. None of that was in my notes yet. The draft was accurate as far as it went, which is exactly the problem, because "as far as it went" is not the same as the inspection I had actually carried out.
The parts you can reasonably delegate
Delegation is the right word here, not abdication. You remain responsible for everything in the signed report. But there are parts of the production process where AI assistance is low-risk and high-value, provided you check the output before sign-off.
Standard narrative for observed conditions. If you have noted "partial slipped tiles to rear pitch, evidence of moss growth, no visible daylight from loft," a well-designed system can produce serviceable draft text describing that condition, its implications, and the standard advice about maintenance or further investigation. The observation is yours; the paragraph is the system's first attempt. Your review confirms it accurately represents what you saw.
Clause selection from a curated library. A library of checked, practice-approved phrases removes a large part of the blank-page problem. AI matching your notes to those phrases is faster than scrolling. The risk is low as long as the library itself is sound. See the article on phrase libraries for RICS surveyors for what a well-constructed library looks like.
Omission flagging. If your notes for a section are thin relative to what that element type normally requires, a system that flags the gap while you are still at the property is worth having. It is a prompted checklist, not a substitute for inspection thoroughness.
Formatting and structure. Getting the standard elements of a Level 2 or Level 3 report into the right order, with the right headings, is mechanical work. Software handles it reliably. This is not a judgement call.
The parts that demand your personal review
Some elements of the report cannot be delegated in any meaningful sense. The AI draft is a starting point, not an answer.
Condition ratings. The RICS Home Survey Standard is explicit that condition ratings reflect the surveyor's professional assessment. No software assigns these on your behalf, and if one purported to, you should not trust it. You enter the rating; the software records it. Review every rating before sign-off against your site notes and photographs.
Risk and significance of defects. Whether a defect is serious enough to recommend a specialist report, and what the likely cost implications are, involves judgement the AI does not have. The drafted text may suggest further investigation; it is your call whether that recommendation is appropriate for the property you inspected.
The overall condition and market context. The summary sections of a Home Survey report carry significant weight with clients. AI-drafted summaries should be read with particular care. They are based only on what you noted; they have no sense of how the property compares to the local market or to what the client is likely to do with it.
Anything the site notes do not cover. AI works from what you gave it. If an important observation is missing from your notes, the draft will not invent it, but it also will not know it is missing. Your professional responsibility includes making sure the report reflects the inspection you actually carried out.
After a few months of using AI-assisted drafting regularly, I noticed I was reading the output too quickly. The prose was fluent, the structure was right, and there was nothing obviously wrong, which made it easy to move on. So I put a simple rule in place: before I accept any section, I read it alongside the photographs from that element, not just my notes. The photographs do not lie about what I actually saw. That habit caught two or three things in the first month alone, observations I had made on site but not yet typed up when the draft ran.
What to check in any RICS home survey software before you rely on it
Surveyors adopting AI-assisted report writing should satisfy themselves on a small number of points before putting their name to output.
Audit trail. The RICS AI standard, as set out in RICS guidance on AI use in professional practice, requires that AI-generated content be identifiable and that the surveyor's review and amendment of it be recorded. Check that the software logs what the AI drafted and what you changed. The audit trail question is worth examining in detail before you commit to any platform.
Source of the phrase library. Ask who wrote the library, when it was last reviewed, and whether it reflects the current Home Survey Standard. A large library of outdated or generic phrases is not an asset.
How easy is the review step? Software that makes review awkward, or that buries the AI-drafted text in a way that makes it easy to sign off without reading carefully, increases rather than decreases your risk. The interface should make review the natural next step after drafting, not an optional extra.
What the system does with your data. Site notes and property photographs are client data. Check the data processing agreement, understand where data is stored, and confirm the provider's security posture before you start uploading sensitive information.
SiteWrite addresses the audit trail point directly, logging AI-drafted text and surveyor amendments separately so the record is clear at sign-off. You can read more about how the platform approaches RICS Home Surveys if that is relevant to your evaluation.
A surveyor who understands this split, what the AI is doing, what it cannot do, and what to check before signing, can use AI assistance confidently. One who treats the draft as the finished article is taking on risk that the software cannot absorb on their behalf.
Frequently asked questions
Can AI software assign condition ratings in a RICS Home Survey?
No current AI in surveying software assigns condition ratings, and it should not. Condition ratings are a professional judgement that belongs to the inspecting surveyor. You enter the rating; the software records and incorporates it into the report. Be cautious of any system that implies otherwise.
If AI drafts part of my report, am I still liable for its content?
Yes, fully. Your signature on the report means you stand behind every word in it, regardless of how it was produced. AI assistance changes the speed and ease of drafting; it does not transfer liability. The RICS position on AI in professional practice is consistent on this point.
How do I know if the AI-drafted text accurately reflects my inspection?
Read it against your site notes and photographs before sign-off. The draft is based entirely on what you gave the system. If your notes were incomplete or ambiguous, the draft will reflect that. The review step is not optional; it is where your professional judgement re-enters the process.
What should I look for in the audit trail of an AI-assisted report?
At minimum: a record of what text the AI generated, what the surveyor changed, and when. This allows you to demonstrate, if challenged, that you reviewed the AI output and made it your own. RICS guidance on AI use in professional practice points to transparency and accountability as the key requirements.
Does using AI-assisted software mean my reports are less defensible?
Not if you use it correctly. A clear audit trail of AI drafts and surveyor amendments can actually strengthen your position by showing the steps you took. The risk comes from treating the draft as final without review, or from using a system that does not log the distinction between AI output and surveyor sign-off.