AINorthern IrelandBusiness

Summarising Consultation Responses With AI: A Method Councils Can Defend

A step-by-step method for councils to summarise public consultation responses with AI: preparing data, theming, checking against the source and reporting in a way you can defend.

Geoffrey Moffett 4 Min Read

Quick Answer

To summarise consultation responses with AI, councils should remove personal data, use an approved tool to suggest themes, have officers check those themes against a sample of the original responses, and report the method openly. AI speeds up sorting and drafting, but officers remain responsible for the analysis and every response must still be considered.

Key Takeaways

  • AI helps sort and summarise; it doesn't replace officers reading and considering responses.
  • Remove personal data before analysis unless the tool is approved for it.
  • Check AI themes against the raw responses, especially minority views.
  • Explain the method in the consultation report.
In This Article

A consultation on a leisure strategy closes with hundreds of free-text responses. Some are a line long, some are four pages. Somebody has to read them, theme them and write a report members can rely on. Summarising consultation responses with AI can save councils days of work, but the method needs to stand up to scrutiny from members, respondents and, potentially, a challenge.

For the wider argument about better consultation, see our post on how AI is changing community dialogue. This one is the practical how-to.

Ground Rules Before Summarising Consultation Responses With AI

  1. Every response is still considered. AI helps organise and summarise; it doesn't decide what matters.
  2. Personal data is protected. Follow your council's data protection approach and check whether a data protection impact assessment is needed.
  3. Officers own the analysis. The report says what officers found, supported by tools.
  4. The method is written down. If asked how responses were analysed, you can explain it.

Step 1: Prepare The Data

  • Export responses into a spreadsheet, one response per row, with a unique ID.
  • Remove or redact names, addresses, email addresses and identifying details unless your tool is approved for personal data.
  • Keep closed-question results separate; they're analysed with normal tools.
  • Note responses from organisations separately if your methodology treats them differently.

Keep the original, unredacted data in your normal secure system.

Step 2: Read A Sample First

Before using AI, an officer reads a random sample, perhaps 30 or 40 responses, and notes the themes they see. This gives you a human baseline to compare the AI's themes against.

Step 3: Ask AI For Themes

In an approved tool, provide batches of anonymised responses with their IDs:

"These are anonymised responses to a council consultation about [topic]. Identify the main themes. For each theme give: a short name, a one-sentence description, the IDs of responses that mention it, and two short representative quotes with their IDs. Also list any responses that raise points not covered by the main themes. Don't infer views that aren't stated."

Asking for IDs is essential. It means every claim can be traced back to real responses.

Step 4: Check The Themes

Now compare.

  • Do the AI's themes match what officers found in their sample?
  • Pick five or ten IDs under each theme and read the originals. Do they actually say that?
  • Read every response flagged as "not covered by the main themes". Minority views often matter most.
  • Look for themes AI may miss: local place names, sarcasm, references to past decisions.

Adjust themes, merge or split them, and record what you changed.

Step 5: Code Responses Consistently

Once themes are agreed, you can ask AI to tag each response against the final theme list. Then check a sample of the tagging, perhaps 10% or more, and correct errors. If accuracy is poor, tag manually or refine the instructions.

Step 6: Draft The Summary

With checked themes and counts, AI can help draft the report section:

"Using these themes, descriptions and counts, draft a plain-English summary for a council committee report. Present views fairly, including minority views. Don't use the word 'most' unless the count supports it. Use the representative quotes provided."

The officer then edits for accuracy and balance. Be careful with numbers: free-text consultation isn't a poll, so describe the strength of views carefully.

Step 7: Explain Your Method

Include a short method note in the report, for example:

"Free-text responses were anonymised and grouped into themes with the help of an AI tool. Officers reviewed the themes against the original responses, checked a sample of the tagging and read all responses that raised issues outside the main themes. All responses were available to officers in full."

That openness supports trust and helps members understand what they're reading.

Closing The Loop With Respondents

AI can also help with the part councils often skip: telling people what happened. Draft a short "you said, we did" summary for the website and social media, in plain English, once decisions are made.

A Made-Up Example: What The Check Caught

Imagine AI themes a set of responses about a park masterplan and reports that "most respondents support the new car park". When officers check the IDs listed under that theme, they find many of those responses actually support a car park but oppose its proposed location, which the AI had grouped together. Officers split the theme in two, and the report presents the picture accurately. Without the check, members could have been given a misleading summary.

Pitfalls To Avoid

  • Treating AI themes as findings without checking. They're a starting point.
  • Losing the trail. Without response IDs, you can't verify anything.
  • Overstating majorities. Free text isn't a representative sample.
  • Ignoring campaigns. Template responses from organised campaigns should be identified and reported transparently, not silently merged.
  • Forgetting equality considerations. Check whether responses from particular groups raise distinct issues relevant to your equality duties.

How Starien Can Help Train Council Teams

Making sense of community feedback was part of our six years leading Coleraine BID's digital delivery, and we've worked with public sector teams who need methods they can stand behind. Our AI Coaching sessions walk policy and engagement officers through this process with practice data, and our Public Relations support helps you report back to the public clearly. Get in touch to find out more.

Frequently Asked Questions

Is it acceptable for councils to use AI to analyse consultation responses?

It can be, as a support tool, if data protection is handled properly, outputs are checked by officers and the method is explained. Decision-makers must still be able to show that responses were properly considered.

Can AI miss important points in consultation responses?

Yes. It can underweight unusual, minority or nuanced points and sometimes misread sarcasm or local references. That's why officers should check themes against the source and read responses flagged as distinct.

Should we tell respondents AI was used?

Being open about the method in the consultation report is good practice and supports trust. Explain what AI did and what people did.

Written By

Geoffrey Moffett

More Articles
More In AI

Back To Top