What AI drafts get right, and wrong, in planning reports

Structure is the part of a planning document that automation handles well, and evidence is the part it cannot touch at all. That is the honest division, and understanding it is what makes AI planning report quality a manageable question rather than a matter of faith. A drafting tool knows what a Transport Statement contains, what order the sections go in, and which policies a change of use engages. It does not know what the traffic count said, whether the bat survey found anything, or which of two defensible readings of a policy your site can sustain.

For a developer buying drafts at volume, the practical consequence is that review time should be budgeted against the evidence sections, not spread evenly across the document.

Why structure is the easy part

Planning documents are conventional by design. An officer reading a Flood Risk Assessment expects site description, flood zone classification, sources of flooding, climate change allowances, sequential and exception tests, finished floor levels and residual risk, roughly in that order, because that is the order the guidance implies and the order every reviewer is used to.

Conventions like that are learnable. So are the policy hooks: a change of use from Class E to a hot food takeaway engages noise, odour, extraction, parking and hours, and any competent drafting will raise all five whether or not the author has seen the site.

What follows is that a draft is usually strongest at the beginning and the end of each section, and weakest in the middle, where the site specific content should be. A section that opens with the right policy framing, closes with the right conclusion, and contains three sentences of generic description in between is the characteristic shape of a document that needs work.

Where the section templates come from

PlanScope's drafts are structured against the section templates for each of its 66 report types, grounded in national policy and local plan text, with citations to policy and flags where facts are missing. The drafting inserts a "[TO BE CONFIRMED]" marker rather than inventing site data, and the output passes a four layer check covering structure, policy, factual grounding and professional standards.

Those markers are the most useful thing in the document. They are a list of the facts the report needs and does not have, which is precisely the list a project manager wants before commissioning survey work. Treat them as a scoping tool rather than as an inconvenience to be deleted.

Report types are also labelled honestly as professional required or professional recommended, with the governing legislation or standard. Read that labelling before the drafting starts, because it tells you which documents can be finished in house and which need a named author.

Evidence is the thing that cannot be generated

Everything in this list has to come from somewhere outside the document:

• Survey results: traffic counts, noise measurements, bat emergence surveys, trial pits, ground gas monitoring, tree schedules.
• Site facts: levels, ground conditions, existing use history, floor areas measured rather than scaled, the position of the original rear wall.
• Consultee positions: what the lead local flood authority said at pre-application, what the highway authority will accept at the access.
• Local policy detail: the specific wording of the adopted plan, the local validation list, the parking standard in the relevant supplementary planning document.
• Anything with a number attached to a real measurement.

A draft that contains a figure nobody in the team can trace is the failure mode to watch for. Check numbers against their source before anything leaves the office, using the approach in our note on factual errors to look for in a drafted planning document.

Councils increasingly ask who prepared a document and against what competence, and the answer needs to be straight. Our note on when a council asks who wrote your report covers how to handle that.

Judgement calls a draft will not make

Four categories, in rising order of difficulty.

Weighting. How much weight to give an emerging local plan policy, a recently adopted neighbourhood plan, or an appeal decision on a nearby site. This is a matter of professional judgement informed by the specific circumstances, and a draft will hedge rather than decide.

Strategy. Whether to argue that a harm does not arise, or to accept it and argue that the benefits outweigh it. Those are different documents and you have to pick one. Arguing both, which is what an unedited draft often does, tells the officer you have not decided.

Concession. What to concede so as to win the point that matters. On a contested scheme the value of a planning statement lies as much in what it declines to fight as in what it argues.

Risk. Whether a marginal position is worth taking given the cost of refusal, the programme, and the authority's recent behaviour on comparable cases. Nothing in a draft knows any of that.

Tone and register on a contested application

A neutral, well organised document is right for a straightforward application and wrong for a contested one.

On a scheme with 40 objections and a likely committee determination, the supporting statement has a second audience: members who will read the officer report and possibly nothing else. That changes the writing. Shorter sentences, the benefits stated early, the objections addressed by name rather than by category, and a clear statement of what the applicant has changed in response to consultation.

Drafts default to the neutral register, which is the right default. Someone has to decide when to depart from it, and that decision is not a drafting task.

There is a related trap. A document that reads as though it was produced without reference to the site is a document an objector will use. Photographs, local place names, the actual bus route number, the name of the school 200 metres away: specificity is not decoration, it is evidence that the author went there.

Deciding what to keep and what to rewrite

A triage that works on a document of any size:

The middle row is where the value sits. Keeping a well organised policy by policy structure and replacing the reasoning inside it is faster than starting from a blank page, and it produces a document an officer can follow.

A structured internal checklist makes this repeatable across a team, and our note on building an internal review checklist for submissions sets out what to include.

Budgeting review time honestly

The mistake developers make is treating a draft as a near-final document that needs proofreading. The realistic split on a typical supporting statement:

1. Reading the draft against the drawings and the site: not skippable, and longer than people expect.
2. Replacing site description and technical content: the largest block of time.
3. Rewriting the policy reasoning to take a position: moderate, and the part that needs the most senior person.
4. Checking every number and citation to source: short, and the one most often skipped.
5. Final read for internal consistency after edits: short, and it catches the contradictions that edits introduce.

Where the time saving is real: the first draft exists, the structure is right, the policy hooks are all present, and nobody spent an afternoon assembling a framework from a previous job. Where it is not real: any document whose substance is measurement.

For report types where a licensed or competent professional is legally required, including Environmental Impact Assessment, Protected Species Survey, Biodiversity Net Gain Assessment, land contamination reporting and fire statements, a draft is a starting point for that professional and never a substitute. The professional directory is the route to a reviewer who can check and sign off a draft, and you can see pricing for the free report, the per-report option and the Professional tier.

This is general guidance, not legal advice. PlanScope provides AI-assisted draft reports for planning applications. Reports should be reviewed before submission. We do not guarantee planning approval. For specialist technical reports, qualified professional input may be required.