AI in tree survey reporting: what it can and cannot do

A tree report is a professional opinion. When it is challenged — after a failure, a claim, or a planning refusal — the question is whether a competent arboriculturist made the assessment, on what evidence, and whether that is documented. Automated text is no defence at all if nobody qualified stood behind it.

That constraint sets the boundary cleanly. Automation is legitimate wherever it handles data the surveyor has already decided, and illegitimate wherever it makes the decision. Every claim about AI in survey software should be tested against that line.

Most of the time lost in survey reporting is not thinking time — it is handling time. The savings that are real, available now, and carry no professional risk:

• Structured capture: recording against defined fields in the field, so there is nothing to transcribe afterwards. This is the single largest saving available to most teams.
• Automatic export in the client's format, rather than reformatting a spreadsheet per client.
• Dashboards and print-ready summaries generated directly from the survey record, so progress reporting is not a separate piece of work.
• Consistency checking: flagging records with a missing mandatory field, an impossible dimension, or a duplicate reference before they reach the client.
• Automatic asset referencing, so two surveyors working the same corridor cannot allocate the same reference.

None of that is glamorous and none of it requires a language model. It is also where almost all the recoverable hours are.

There are three uses that stand up:

• Narrative drafting from structured data. The surveyor has already recorded species, condition, defects and recommendation; generating the prose paragraphs from those fields removes repetitive writing without touching the judgement.
• Image triage at volume. On a programme with tens of thousands of photographs, flagging images that appear to show significant deadwood or crown loss helps prioritise which get reviewed first. A prompt to look, not a conclusion.
• Pattern surfacing across a dataset. Highlighting that decline is concentrated on one section of corridor, or that one species is over-represented in urgent works, is analysis a person could do but rarely has time for.

In all three the output is an input to a professional, reviewed before it goes anywhere near a client.

• Assigning a condition class or ash dieback class from a photograph. Season, angle, light and species variation all move the result, and the failure mode is a missed defect on a tree beside a carriageway.
• Generating a works recommendation. That is a risk judgement about a specific target, not a text-generation task.
• Producing a report that ships without a named surveyor reviewing every substantive statement in it.
• Inventing anything the survey did not record. If a field was left blank in the field, the report says so — it does not fill the gap plausibly.

• Is the AI feature available on the plan I am buying today, or on a roadmap?
• What exactly does it produce — a draft for review, or output that reaches the client unreviewed?
• Where does my survey data go, and is it used to train anything?
• Can I see the underlying record behind any generated statement?
• If the generated text is wrong, whose professional indemnity covers it?

The first question filters most of the market. A great deal of arboricultural AI marketing describes features that are not yet purchasable.

We would rather be straight about this. Arbrium's reporting today is built on structured data rather than generated text: surveys captured against defined templates, exports in the client's own format, insights dashboards and print-ready summaries produced directly from the record. That is what removes hours from a programme now.

AI Copilot — narrative drafting and assisted evidence review — is on our roadmap and we do not sell it as available. When it ships it will produce drafts for a named surveyor to review and sign, because that is the only version of it that survives a liability question.

For the survey methods behind the data, see our guides to tree safety inspections and drone tree surveys.