Can AI Help Write an Arborist Report? What Stays With the Arborist

AI can take an arborist's site notes and photo captions and arrange them into a clear report draft: summary, site description, tree-by-tree findings and a recommendations table. The measurements, condition and risk ratings, and every recommendation stay with the arborist, who checks each line and signs.

The trap is fluency. A draft that reads well can hide a changed fact, such as "possible decay" becoming "decayed" or an invented trunk diameter. The method below keeps the AI to wording and layout, and gives you a way to catch the places where it slipped.

Follow me on Instagram@sagnikteaches

Where the judgement sits in a tree report

Sort the report into three layers before you hand anything to an AI tool. The first layer is observation: what you measured and saw on site. The second is assessment: what those observations mean for the tree's condition and the risk. The third is communication: how the findings are laid out for a client, an insurer or a planning officer.

Connect on LinkedInSagnik Bhattacharya

AI belongs in the third layer. It is good at turning a short note into a tidy paragraph, keeping a consistent order, and writing a plain-English summary for a non-expert client. It is poor at the first two layers because it has not stood at the tree. It cannot see the included bark union, smell the fungal fruiting body, or feel the soil heave.

Subscribe on YouTube@sagnikteaches

If you work to a formal risk method, such as the one taught for the ISA's Tree Risk Assessment Qualification (TRAQ), the rating comes from your trained assessment of likelihood and consequence. Nothing in this tutorial changes that.

Start with notes the AI can use

A draft is only as good as the notes behind it. The commonest cause of a poor AI draft is a note that was clear to the person who wrote it and ambiguous to anyone else. Use one fixed note format per tree, whether you type it, dictate it into a voice-notes app or fill in a form on a tablet.

Tree ID: T07
Species: common oak (your call, not a guess)
Height m: 16  Stem diameter at 1.5 m (cm): 74  Crown spread N/E/S/W (m): 6/5/7/5
Age class: mature
Condition: fair. Dieback ~15% of crown, upper south side. Cavity at
  base of main limb, east side, approx 10 cm wide, depth not probed.
Defects: included bark union at 4 m, no cracking seen.
Targets: garden shed 8 m SW, footpath 12 m N.
Photos: P112, P113, P114
Recommendation (mine): crown clean, remove deadwood over 25 mm,
  re-inspect cavity in 12 months.
Not assessed: roots below ground, upper crown (no climbing).

Two habits improve everything that follows. The first is a line for what you did not assess, since a report's limitations come from there. The second is writing your own recommendation per tree in your own words, so the AI never has to originate advice.

The drafting prompt, with the guardrails in it

A good prompt tells the assistant exactly what it may not do. Paste your notes below it.

You are formatting a tree report for an arborist who will check and sign it.
Use ONLY the facts in the notes below. Rules:
1. Do not add any measurement, species, age, defect or recommendation
   that is not in the notes.
2. Keep my hedges: if I wrote "possible", "approx" or "not assessed",
   keep that word.
3. If a field is missing, write [MISSING: field name] instead of guessing.
4. Do not give any risk rating unless I wrote one. Do not change one I wrote.
5. Output, in this order: (a) one-paragraph summary in plain English,
   (b) a table of trees with ID, species, height, stem diameter, condition,
   (c) a short paragraph per tree, (d) recommendations grouped by
   priority exactly as I wrote them, (e) a list of everything not assessed.
6. At the end, list every place where you changed the wording of a defect.

NOTES:
[paste notes]

Rule 6 is the useful one. It forces the assistant to point at its own rewrites, which is where subtle changes of meaning happen. You then read those lines first.

Worked numbers: a 14-tree survey before an extension

Take a solo arborist surveying a residential plot with 14 trees before a small extension, for a homeowner. Everything here is illustrative.

On site. Walking the plot takes 90 minutes. The arborist dictates each tree into a phone using the note format above, 14 notes of about 60 to 90 words each, and takes 3 to 6 photos per tree.

Writing it up the old way. Transcribing the notes, setting the table, writing 14 paragraphs, the summary and the limitations, takes around five hours across an evening and a morning.

Writing it up with a draft. The notes go into the prompt above. The assistant returns the table, the 14 paragraphs and the summary in a couple of minutes. The arborist then spends 20 minutes on the "changed wording" list, 60 minutes reading every paragraph against the notes and photos, 20 minutes adjusting the summary, and 15 minutes on the limitations and sign-off. That is about two hours. The saving is roughly three hours on this job.

The saving only exists because the checking time is counted. Cut the checking and you have saved a further hour and bought a liability. Value the three hours at your own day rate and compare that to a business-plan AI seat; the sums are in your favour only if the checks stay.

What a good draft changes and what it must not

Here is one tree from the earlier notes, as the assistant might return it. The sample is illustrative.

T07 common oak. A mature oak, 16 m tall with a stem diameter of 74 cm at 1.5 m and a crown spread of about 5 to 7 m in each direction. Condition is fair. Roughly 15% dieback is visible in the upper south side of the crown. A cavity, approximately 10 cm wide, sits at the base of the main limb on the east side; its depth was not probed. An included bark union is present at about 4 m with no cracking seen. The nearest targets are a garden shed 8 m to the south-west and a footpath 12 m to the north.

That is faithful. Compare a version that would need fixing: "The oak has significant decay at the base of the main limb, and the included bark union is a likely failure point." Here "cavity, depth not probed" became "significant decay" and a union with no cracking became a "likely failure point". Neither claim is in the notes. This is the typical failure: the assistant tells a more dramatic story than you did. Your rule 2 and rule 6 exist to catch it.

Before and after: recommendation wording

Recommendations are where clients act, so wording matters. Your note might read: "crown clean, remove deadwood over 25 mm, re-inspect cavity in 12 months." An illustrative draft can come back looking client-ready: "Priority 2 (within 6 months): crown clean, removing dead wood over 25 mm in diameter. Priority 3 (monitor): re-inspect the basal cavity in 12 months."

It reads well, but "Priority 2", "Priority 3" and "within 6 months" came from nowhere in your notes. The faithful version keeps your three elements and adds only layout: "Crown clean, removing dead wood over 25 mm in diameter. Re-inspect the basal cavity in 12 months." If your firm uses priority bands and timescales, put them in your own note format so they come from you. If a draft invents a timescale, delete it. A timescale is a recommendation, and recommendations are yours.

Writing the plain-English client summary

The summary is where AI saves the most effort and invites the most risk, because it compresses. Give it a narrow brief: "Write a summary of 120 to 160 words for a homeowner with no tree knowledge. Mention only trees where I recommended work. Do not say any tree is safe or unsafe." The last instruction matters. Words such as "safe" claim a conclusion, and the report's conclusions are your assessment.

Read the summary against the recommendations table, not against your memory. Count: if the table has three trees needing work and the summary mentions four, one of them is wrong. For a related photo-to-report process in another trade, see roof inspection reports from photos; the checking habits carry across.

Using photos without letting them replace your eyes

Photos help the draft in two ways: you can attach captions, and the assistant can describe what is visible. Use the first and be cautious about the second. A caption you wrote ("P113: cavity east side of main limb, base, ruler for scale") goes into the report as written. An AI-generated description of the same photo might call a shadow a crack.

If you do ask a tool to describe a photo, treat it as a second reader, not a witness. Ask: "List anything visible in this photo that my caption does not mention." Then decide for yourself whether each item is real. Anything you cannot confirm on site stays out of the report or goes in as "to be confirmed on re-inspection". For the capture side, turning on-site voice notes into job sheets covers dictation habits that make the raw material cleaner.

A review checklist before you sign

• Every figure in the table matches your notes: height, diameter, spread, count of trees.
• No species, age or defect appears that is not in your notes.
• Each "possible", "approx" and "not assessed" from your notes survived.
• Any risk rating is exactly the one you chose.
• Each recommendation matches your wording, with no added timescale or priority.
• The summary mentions the same trees as the recommendations table.
• The "not assessed" list includes roots, upper crown and any tree you could not reach.
• Standard limitations text is yours, copied from your template, not newly written.
• Client name, address and any other personal details are put back in your own document, not left as placeholders.

Another firm, another report type: a pre-purchase tree check

A smaller job shows the method flexes. Suppose a buyer wants a quick check of three mature trees in a garden before the purchase goes through. The report is two pages. You still use the fixed note format, but the output prompt asks for a short narrative per tree, an overall "matters the buyer should raise with the seller" paragraph and a limitations section.

The rule that matters most here is the limitation text. A pre-purchase check is a visual inspection from ground level on one day. Put that in your standard wording and instruct the assistant: "Insert the limitations text exactly as supplied. Do not paraphrase it." Wording changes in limitations can alter what you have promised, so keep that paragraph word for word.

A mistake that looked like a typing slip: two trees merged

A realistic failure from a multi-tree survey: two neighbouring trees, T11 and T12, both ash, with similar heights and notes dictated one after the other. The transcription ran the end of T11's note into the start of T12's, and the draft produced one paragraph that gave T11's dieback to T12 and left T11 with no defects at all. The table looked right because heights were different. The paragraph read smoothly. Nobody would have noticed without reading against the notes.

Two habits would have caught it. First, the prompt rule "one paragraph per tree ID, and print the tree ID at the start" makes a merge visible as a missing ID. Second, count the trees: the table had 14 rows and the paragraphs numbered 13. A simple count of IDs in the notes, the table and the narrative is a thirty-second check that finds this class of error every time.

When the notes are thin: what to do with the gaps

Not every job has full notes. Suppose you only wrote "T03 beech, poor, remove" on a busy day. An assistant given that line and told to write a paragraph will fill the gaps with plausible filler, such as describing a "significant structural defect" you never recorded. That is why rule 3 asks for [MISSING: field name] markers.

A thin note should produce a thin paragraph with markers: "T03 beech. Condition: poor. Recommendation: removal. [MISSING: height] [MISSING: stem diameter] [MISSING: defects observed] [MISSING: targets]." That output is useful because it tells you exactly what to fill in from memory, photos or a return visit. If the draft comes back looking complete from a thin note, the assistant has invented content, and you should discard it and tighten the prompt.

A quick sum on cost and where the time goes

Illustrative: ChatGPT Business and Claude Team each require at least two seats, so a solo operator is usually better on an individual plan at about $20 a month; check current list prices. Say you write 6 reports a month and each one saves about two hours after checks. That is 12 hours. If you value an hour at $40, the saving is $480 against a tool cost of about $20 to $50. The cost is not the question. The question is whether you can keep the review disciplined when you are busy, because the saving disappears the first time a changed fact reaches a client.

A second cost people forget is template upkeep. Plan an hour a quarter to revisit your prompt, your standard limitations text and the checklist, particularly after your insurer, professional body or clients change what they expect in a report.

Adapting the template for different report types

A tree survey for a development

The report's audience is a planning officer who wants a consistent table, categories and constraints. Put your category scheme into the note format, so the assistant formats it and never assigns it. Ask for the table in a fixed column order and keep any plan or drawing references as plain text you supplied.

A hazard or risk inspection for a landowner

Here the readers care about what could fall on whom. The assistant should mention only the targets you wrote down. Add a rule: "Do not describe any tree as safe, unsafe, dangerous or healthy unless I used that word." A distinction between "no defects observed" and "safe" is the sort of thing a report can be challenged on later.

A tree-preservation or consent statement

If your report supports an application to work on protected trees, the statement of reasons must be yours. Let the assistant tidy the structure and grammar, but have your own wording for the justification, and read the final text against the rules of the body you are applying to.

A short script for the client conversation

Clients sometimes ask whether the report was written by a computer. An honest answer you can use: "I inspected every tree myself and wrote the findings. I use a drafting tool to help with layout and wording, and I check every line before I sign it." Adapt it to match your terms of engagement, and only say it if it is true of your process.

Handling client data and your insurer

Treat site notes as client information. Remove names, phone numbers and the full address before pasting, and use a site code. Consumer chat accounts let you switch off use of your chats for model training, and business plans generally don't train on your content by default, but check the settings and terms of whichever tool you use rather than assuming.

It is also worth a short conversation with your professional indemnity insurer. Ask whether drafting with AI affects cover, and whether they want a note in your terms of engagement. The answer varies by insurer and policy, so get it in writing. That is a business question, not a technical one, and no tutorial can answer it for you.

Building the template once so the work repeats

Save the prompt, the note format and the checklist in one place, such as a shared document or a Project in a business chat tool. After three jobs you will know what the assistant tends to change, and you can add a rule to the prompt for each. After ten jobs the prompt is a tool you trust because you built the guardrails from its own mistakes.

A good first month looks like this: use the routine on two small jobs alongside your normal method, compare the results, and note the minutes. Move to using it on every report only if the comparison shows both time saved and no change of meaning. To see where this fits among the other desk jobs in a tree business, read seven jobs to take off the owner's desk, and to carry report recommendations into pricing, the estimate template tutorial picks up where this one ends.

What to keep doing by hand

Walk the site yourself. Probe the cavity yourself. Choose the rating yourself. Sign it yourself. Everything else in the report can be tidier and quicker with help, but the reason a client pays for an arborist is a trained person has looked at the tree and stands behind the conclusion. AI that protects that, by saving you the typing, is worth using; AI that quietly replaces it is a risk to your name.

Questions arborists ask about AI-drafted tree reports

Can AI identify tree species from a photo?

It can suggest a species, and often gets common ones right, but it can be confidently wrong on similar species or on trees out of leaf. Treat a suggestion as a prompt to check, never as the identification. In a report you sign, species is your call, recorded from your own inspection.

Should I tell the client I used AI to draft the report?

Check your professional body's guidance and your terms of engagement. Telling the client is the safer course whenever a tool touches their data, and the report stays your work because you wrote the findings and signed it. Your insurer may have a view too, so ask before you start.

Can AI give a tree a risk rating?

It should not. A risk rating rests on what you saw, the target, the likelihood of failure and the consequences, judged by someone trained and accountable for it. Let AI format and word your rating, never choose it.

Is it safe to paste client site details into a chat tool?

Strip names, phone numbers and full addresses before pasting, and use a business plan with a no-training default if you can. Use a site reference in place of the address, and put identifying details back in your own document afterwards.

Further reads

• Tree Surgeon Risk Assessment Template: AI Drafts, You Sign Off — The job-site risk assessment, a different document from this report.
• Mould Inspection Report Template: Findings, Photos and Limits With AI — Findings, photos and limitations text in another inspection trade.
• Structural Engineer Inspection Report: Photos and Notes to a Draft With AI — Same sign-off principle for engineers' inspection reports.
• Tree Service Contract With AI: Access, Damage and Debris Clauses to Verify — Access, damage and debris clauses to verify.
• Tree Service Proposal Template: Commercial and Storm Cleanup Bids With AI — Bid wording when report recommendations become commercial work.
• ChatGPT for Landscape Architects: Planting Notes, Reports and Emails — A checking routine for using ChatGPT on planting notes, reports and client emails: it drafts the words, you choose and verify every plant.
• AI Voice Assistant for Tree Surgeons: What It Can Book and Quote — What an AI voice assistant can capture, book and quote for a tree surgery firm, plus photo requests, access notes and storm-damage calls.
• Using AI in Environmental Consulting: Draft Site Report Sections — Prepare site-report sections from controlled notes and lab results, with a traceable check of every number and conclusion.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ISA credentials pages (Tree Risk Assessment Qualification), OpenAI and Anthropic plan and privacy pages; the worked figures and sample notes are illustrative.