Supervising junior staff who start from a drafted report

Supervising junior planners who work from AI drafts is a different job from supervising junior planners who work from a blank page. The risk has moved. A graduate writing a planning statement from scratch produces something visibly incomplete, and you can see where the thinking stopped. A graduate editing a drafted report hands you something that reads like finished work, with every heading present and every policy cited, and the gaps are invisible until an officer or an inspector finds them.

That is the supervision problem in one sentence. Structural completeness is now free. Evidential completeness is not, and it never was.

What a first draft teaches and what it hides

A good draft gives a junior author the shape of the document: the sections a planning statement needs, the order a heritage assessment runs in, the policy hooks that belong in each part. That is genuine learning, and it happens faster than it used to.

What the draft hides is the reason each section exists. A paragraph explaining how the proposal relates to the settlement boundary is correct in form whether or not anyone has looked at the proposals map. The author who has never had to find that boundary does not know that finding it is the work.

Three specific blind spots appear repeatedly in supervised drafts:

• Policy references that are correct in substance but quoted from a superseded edition. National policy has been reissued more than once in recent years, with renumbering and restructuring, and a draft trained on older material will reproduce old numbering.
• Site facts stated with confidence where the draft inserted a flag and the author resolved it by deletion rather than by research.
• Judgements presented as findings. A sentence saying the proposal would not harm the character of the area is an opinion that needs a reason, and a junior author will often leave it as an assertion because the draft did.

PlanScope's drafting inserts a marker wherever a site-specific fact is missing rather than inventing one. That marker is the supervision tool, and the first rule for any supervised author should be that a marker is resolved by evidence or by an explicit statement that the fact is unknown, never by quietly removing it.

The competence obligation under professional codes

Chartered practice carries a competence duty. The RTPI's code requires members to act with competence and not to undertake work outside their competence, and comparable obligations apply to chartered surveyors and other chartered professionals.

That duty has two consequences in this context. It applies to the supervisor as much as to the author, because signing or submitting a document adopts it. And it is not satisfied by the fact that a tool produced the first draft, since the tool holds no professional status and carries no liability.

Which means the file needs to show who checked what. Not as a defensive exercise, but because a competence obligation you cannot evidence is one you cannot rely on if a client or a professional body asks.

For report types where a licensed or competent professional is legally required, the position is stricter still. A Biodiversity Net Gain Assessment under the Environment Act 2021 regime expects a competent person for the statutory metric. A Protected Species Survey needs a licensed ecologist. An Environmental Impact Assessment must be prepared by competent experts under the 2017 Regulations. No amount of internal supervision substitutes for those requirements, and PlanScope's drafts are a starting point for that professional rather than a replacement. Where you do not hold the competence in house, the professional directory is the route to a reviewer who does.

A two tier review: facts, then judgement

The most useful change you can make to a review process is to split it in two and do them in order.

The first pass checks facts only. Nothing else. Is the address right, is the site area right, is the use class right, are the policy numbers current, do the figures in the text match the figures on the drawings, is every reference to a document one that exists. This pass is mechanical and it can be delegated, checklisted and timed.

The second pass checks judgement. Does the planning balance in the document actually follow from the facts established in the first pass. Is each conclusion supported. Where the document asserts no harm, has someone been to the site or looked at the photographs. This pass cannot be delegated, because it is the professional judgement the signature attests to.

Running them together is what produces the familiar failure: an hour spent arguing about a policy reference while a paragraph asserting acceptability goes unexamined.

Making the author verify rather than tidy

Junior authors editing a draft default to improving the prose. It feels productive and it is the wrong task.

Four habits redirect the effort:

1. Ask for a source list alongside the draft. Every factual statement gets a line in a separate schedule naming where the fact came from, with a date. Statements with no source are flagged, not deleted.
2. Require the proposals map extract, the constraints printout and the site photographs to be in the file before the draft is reviewed. If the author has not seen them, the review is premature.
3. Ask the author to mark, in the document, the three sentences they are least confident about. This takes two minutes and it is remarkably accurate.
4. Review the evidence first and the document second, so you form a view before you read the argument.

That last habit is the single biggest change. A supervisor who reads the drafted document first is primed to accept its framing, and will spend the review checking the document against itself.

Comment conventions that leave an audit trail

Conventions only work if they are short enough to be used. Three categories cover most cases: a query needing an answer, an instruction to change something, and a note recording that something was checked and accepted.

Keep the record in a form that survives regeneration. Where sections of a report are regenerated after a change, comments anchored to the old text disappear, which is a practical problem rather than a theoretical one. The approach in version control when report sections are regenerated matters here: keep the query schedule as a separate document keyed to section headings rather than to paragraph numbers.

What the file should contain when the job closes: the drafted version, the evidence pack, the query schedule with answers, the final document, and a one-line record of who signed and on what date. That is five items, and assembling them takes less time than reconstructing them a year later when the appeal comes in.

Who signs, and what signing means

Someone has to own the document. In practice that should be the person who carried out the judgement pass, and they should be named.

Three rules make this work. The signatory holds the competence relevant to the document, or has had the specialist parts reviewed by someone who does. The signatory has seen the evidence pack, not only the text. And the signatory is willing to defend each conclusion at a hearing, which is a useful test to apply before signing rather than after.

Clients and committees increasingly ask how a document was produced. Being straightforward about it is easier than being evasive, and the approach in explaining an AI-assisted draft to a client or a committee works because it separates what the draft did from what the professional did.

The training value in the checking itself

There is a worry that juniors who start from drafts never learn to write. The worry is reasonable but the diagnosis is wrong. The skill that matters is not composition, it is knowing which statements need evidence and which do not, and checking a draft against evidence teaches exactly that.

What does not teach it is asking a junior to proofread. Give them the fact-checking pass with a checklist and a deadline, and ask them to explain, afterwards, which statements they could not verify and why. The second question is the lesson.

One honest uncertainty worth naming. Nobody yet knows what a cohort trained this way will be like at the judgement pass in five years, because the judgement pass has always been learned by doing the fact pass badly and being corrected. If you shorten that apprenticeship, something has to replace it, and site visits are the obvious candidate. The pattern that drafts get the structure right and the evidence wrong is a reasonable summary of what juniors should be taught to expect.

See PlanScope pricing for the Professional tier if your practice drafts supporting statements regularly and needs a saved project library and your own branding.

This is general guidance, not legal advice; check professional obligations with your institute and the position with your local planning authority or a qualified professional before relying on it. PlanScope provides AI-assisted draft reports for planning applications. Reports should be reviewed before submission. We do not guarantee planning approval.