Run it in six steps: agree confidentiality and rater selection, send a short question set, collect responses with a minimum of three per group, clean the comments, let AI propose themes that you then check line by line against the raw comments, and build the report around those themes. Never let AI see names or decide what matters.
Trust comes from two things that software feature lists rarely show: raters who believe they cannot be identified, and a client who can see where every theme came from. Both are process choices made before any AI tool opens, which is why the steps below start with contracting and end with a debrief.
A worked example runs through the whole process: a coach, a head of operations at a 60-person company, and 12 people invited to give feedback. Every figure is an illustration.
AI helps in exactly one stage, theme coding, and even there it is a first pass. It saves typing, not reading: with 27 written comments, an honest check takes about 25 minutes, and that check is where the report earns its credibility.
Contracting: what the client and raters are told first
Four decisions go into the coaching agreement or a one-page 360 brief, and the client signs it before any invitation leaves your outbox.
- Purpose. Development only, never pay or promotion decisions. The client's manager or sponsor agrees to that in writing and knows what they will see, usually the themed summary rather than raw comments.
- Who the raters are. The client proposes 10 to 14 names and you check the balance with the sponsor: one manager, four or five peers, four or five direct reports and one to three others. Nobody is on the list only because they are friendly.
- Anonymity rules. Scores appear for a group only when it has three or more responses, the manager is shown separately because their identity is known, comments are edited for identifying detail, and nobody tracks who wrote what.
- Data handling, AI included. Who processes the comments, that names are removed first, which kind of plan you use and the date everything is deleted.
In the example, the client proposes 12 raters: one manager, five peers, five direct reports and one external partner. After the sponsor's review, one peer is swapped for a colleague from another department, because all five peers worked on the same project. That leaves four peers and two others. An illustrative invitation that carries the rules to raters:
Hi [first name], [client name] has asked me, as their coach, to gather feedback from people they work with. It is 18 short statements and three questions, about 12 minutes, open until [date]. Your answers are anonymous: scores are shown only for groups of three or more, and I edit comments to remove anything that could identify you. I read every comment myself. I use an AI assistant [on a business plan where the vendor does not train on my data] to help group comments into themes, after names and identifying details are removed, and I delete everything on [date]. Link: [link]. Thank you, [your name]
Fill the bracketed plan line only with something you have checked. AI note takers and confidentiality for executive coaches covers the same promise for session recordings and shows how coaches word it for clients.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA question set that survives ten raters
Eighteen rated statements and three open questions fit in about 12 minutes, which is about as long as busy raters will give. Use a five-point scale from "rarely" to "almost always", and add "cannot say" to every item so raters are not forced to guess about behaviour they have not seen.
| Theme (3 statements each) | Sample statement, rated 1 to 5 |
|---|---|
| Strategy and priorities | Makes clear which priorities come first when demands collide |
| Communication | Keeps me informed of changes that affect my work before I hear them elsewhere |
| Delegation and development | Lets people finish work they have been given, without taking it back |
| Decision-making | Makes timely decisions and explains the reasons |
| Collaboration | Shares context willingly with other teams |
| Handling pressure | Stays steady when deadlines or stakes rise |
Each statement describes one observable behaviour. "Is a good communicator" asks the rater to judge; "keeps me informed before I hear it elsewhere" asks what they saw. The three open questions are: what should [name] keep doing, what should [name] do more of or differently, and what else would help [name] be more effective. Resist adding a fourth, because comments take the longest to write and are where anonymity is hardest to protect.
A coach who already uses client questionnaires can adapt the same method; building and analysing a client questionnaire with AI covers writing items that produce usable answers.
Anonymity thresholds and form settings
A common rule of thumb is at least three responses in a group before its scores appear on their own. A group below that is merged with the nearest group, and a single manager is shown alone because everyone knows who they are. In the example, 10 of the 12 invited people respond:
| Group | Invited | Responded | Shown as |
|---|---|---|---|
| Manager | 1 | 1 | Manager (known by design) |
| Peers | 4 | 3 | Peers and others, 5 responses |
| Others | 2 | 2 | Merged into peers, because 2 is below 3 |
| Direct reports | 5 | 4 | Direct reports |
Three form settings matter if you use Google Forms. Make sure the "Collect email addresses" setting is off; Google's help describes switching it on as a step you take, but confirm it on every form you send. The "Limit to 1 response" setting forces sign-in to a Google account, and although the help page says usernames are not recorded unless you also collect email addresses, sign-in is a trust risk in itself, so skip it. And keep the option that shows respondents a summary of responses switched off, because Google's help page says those summaries, full-text answers included, are visible to anyone who can respond. Check any form tool by submitting a test response and reading the export for anything that identifies you as the sender.
Chasing without identifying
You cannot tell raters apart, so you chase everyone. Send a generic reminder on day 3, day 7 and day 10 ("if you have not yet completed it, the link closes on [date]"), and watch only the count per group. If peers sit at 1 of 4 on day 10, ask the client to nudge that group as a whole. Never ask individuals whether they have done it, because the replies tell you who has not, and by elimination who has.
Cleaning comments so the writer cannot be recognised
Raters rarely sign comments, yet a comment can still point at its author: a named project, a dated incident, a phrase they use in meetings. Cleaning means keeping the meaning and removing the pointer. Do it yourself before any AI tool sees the text, and give every comment an ID and a group tag such as C07 [Peers].
Three illustrative before and after pairs:
- Before: "In the March launch week you stayed until midnight redoing my slides." After: "On a recent deadline, took over a piece of my work and redid it."
- Before: "Unlike [name of previous boss], you never shout." After: "Stays calm when things go wrong."
- Before: "As the only person in Finance who reports to you, I rarely get an update." After: "Updates reach me later than they should."
The third pair is the hard case: the original reveals that the writer is the only finance direct report, so the group tag and the detail together identify them. Remove the detail, and if a group has one such person the group should not be reported on its own. How to anonymise client data before you paste it into AI has a longer checklist, including dates and job titles that identify a person without a name.
Theme coding with AI, then a human check
Paste only the cleaned comments, with their IDs and group tags, into an assistant on a plan whose data terms you have accepted. The prompt separates finding themes from assigning comments, because asking for both at once invites invention.
You are helping a coach code anonymised 360 feedback comments.
Each line is: ID [group] comment.
Step 1. List up to 6 candidate themes, each with a one-line definition.
Mark each as strength or development area.
Step 2. Assign each comment to at most two themes. Show a table:
theme, comment IDs, number of comments, groups represented.
Step 3. For each theme give up to two quotations, copied word for word
with the comment ID. Never paraphrase inside quotation marks.
Rules: do not guess who wrote any comment. Do not infer personality
traits. Mark any theme supported by only one comment as "single voice".
List comments that fit no theme separately. State how many comments you
read.
C01 [Directs] Makes decisions quickly and always explains why.
C02 [Directs] Steps in and redoes work instead of letting us finish it.
C03 [Directs] I usually hear about priority changes second-hand.
C04 [Directs] Trusts us with client work, then checks every detail again.
C05 [Peers] Calm in difficult meetings; people settle when they speak.
C06 [Peers] Slow to share context with other departments.
C07 [Peers] Cuts through the noise and makes the call.
C08 [Manager] Strong operator. Needs to build a successor.An illustrative first result for those eight comments:
| Theme | Type | Comments | Groups | My check |
|---|---|---|---|---|
| Decisive and clear | Strength | C01, C07 (2) | Directs, Peers | Sound; two groups agree |
| Takes work back or re-checks it | Development | C02, C04, C08 (3) | Directs, Manager | C08 is about succession, not re-checking; move it |
| Information does not flow | Development | C03, C06 (2) | Directs, Peers | Sound; two groups agree |
| Calm under pressure | Strength | C05 (1) | Peers | Single voice; keep it out of the headline themes |
Here the human check earns its keep. In the same run, the assistant had grouped C08 under "takes work back" because both comments are about developing others, and it had offered the quotation "always dependable under pressure" for the calm theme. That sentence is not in the file. The comment says "Calm in difficult meetings", and a quotation that is only close would be the first thing a sceptical client notices. The check has five parts:
- Read all 27 comments yourself, in the order raters gave them, before reading the AI's themes. Otherwise you will read them through its labels.
- Recount each theme by hand from the coding table. With ten raters, report counts such as "4 of 10 raters mention it", never percentages.
- Search the comment file for every quotation, word for word. If the search finds nothing, it is not a quotation.
- Keep a theme in the headline list only if at least two raters, ideally from two groups, support it. Anything else is a single voice and goes in a separate short section, or nowhere.
- Look for what the AI left out: the rare comment that is specific, fair and uncomfortable. Those are often the ones worth a conversation in the debrief.
A comment that alleges misconduct, harassment or something illegal does not belong in a theme. Set it aside and handle it under your coaching agreement and professional code, as contracted at the start.
Scores: self against others
Theme scores are the mean of each theme's three statements. Compare the client's self-rating with the mean of everyone else, weighted by the number of responses, and note the gap. In the example, all others means the manager (1), peers and others (5) and direct reports (4), ten raters in total.
| Theme | Self | Manager | Peers and others | Direct reports | All others | Gap (self minus others) |
|---|---|---|---|---|---|---|
| Delegation and development | 4.5 | 3.3 | 3.8 | 2.5 | 3.23 | +1.27 |
| Communication | 4.3 | 3.7 | 4.0 | 3.2 | 3.65 | +0.65 |
| Strategy and priorities | 4.0 | 4.0 | 4.2 | 3.8 | 4.02 | -0.02 |
| Decision-making | 3.8 | 4.3 | 4.4 | 4.0 | 4.23 | -0.43 |
The all-others figure for delegation is (3.3 + 5 × 3.8 + 4 × 2.5) ÷ 10 = 3.23, so the gap is 4.5 minus 3.23, or +1.27. The pattern reads as a blind spot, which matches the "takes work back" theme: the client rates themselves highly on letting people finish work, and the direct reports, who see it daily, rate it lowest. Decision-making shows the opposite, a hidden strength, where others rate the client higher than the client does. Present gaps as questions ("what do you make of this?"), because ten raters give you a pattern, not a measurement.
A report layout the client can trust
Six to eight pages is enough, and a client who can trace each claim to its source will trust the rest.
- One-page summary. Three strengths and three development themes, each in a sentence, with the rater count behind it.
- Scores page. The table above, with group sizes and a note on the merging rule.
- Strength themes. A short description, two or three verbatim comments tagged by group, never by person.
- Development themes. The same layout, with a question for the client to consider after each.
- Single voices. Optional, clearly marked as one person's view.
- How this report was made. Number of raters and responses, the anonymity rule, the fact that comments were cleaned, that AI was used to propose themes and that you checked every theme and quotation against the raw comments, plus the deletion date.
The last section is the one most coaches skip, and it is the cheapest trust you can buy. A client who learns of the AI step from your report, in your words, takes it better than one who learns of it from a rater.
A 90-minute debrief, minute by minute
Send the report 48 hours ahead so the first reaction happens in private, and tell the client to read it once and mark only what surprises them.
| Minutes | What happens | Your job |
|---|---|---|
| 0 to 10 | Frame the session | Restate purpose, confidentiality and how to read a 360: patterns, not verdicts |
| 10 to 25 | First reactions | Ask what stood out and what they expected. Say nothing about interpretation yet |
| 25 to 50 | Scores and gaps | Take the largest gap and one hidden strength; ask for the client's explanation of each |
| 50 to 70 | Comments | Read two or three comments aloud per theme and ask what the client hears in them |
| 70 to 85 | Choose priorities | Agree one or two themes to work on and one small experiment for each |
| 85 to 90 | Next steps | Fix the session date for progress and a short re-pulse in about six months |
Start with strengths, because a client who hears only gaps defends instead of listening. In the example, the head of operations opens by saying the delegation gap is unfair; reading the three direct-report comments aloud, without comment, turns the discussion within minutes, and the experiment they agree is to hand over one named piece of client work with a written definition of done and no mid-way re-checks.
Tool options and what they cost
The coach's steps above work with any of these. Prices are the vendors' own listings in November 2026.
| Route | What you get | Listed cost |
|---|---|---|
| Form, spreadsheet and an AI assistant | Full control of questions and anonymity rules; you build the charts and the report | A form builder you may already have, such as Google Forms, plus an assistant on a business plan such as ChatGPT Business: Standard seats $20 a month billed annually, two seats minimum |
| Spidergap | 360 platform with a questionnaire template, anonymised feedback, reports and a development plan; its pricing page names no AI feature | Starter $1,099 a year including 10 feedback recipients, extra recipients $89 each; Pro $2,199 a year, extra recipients $59 each |
| Your360.ai | Voice AI interviews with raters and an AI-coached debrief for the participant | Individual $549; Team $3,995 for up to 8 recipients with up to 12 feedback providers each, $299 per extra recipient |
| smart360 | Text and voice feedback with AI-powered analysis | No price shown on its page |
A quick sum for a coach who runs six 360s a year. Spidergap Starter works out at $1,099 ÷ 6 = about $183 per client, falling to $110 per client at ten. Your360's Individual price across six clients is 6 × $549 = $3,294, while its Team plan is $3,995 ÷ 8, about $499 per recipient. The two-seat assistant plan costs $40 × 12 = $480 a year and covers every client, though not the chart-making.
The AI-led options change what you sell. Your360 replaces the human debrief with an AI one, which suits an organisation buying at volume and does not suit a coach whose value is the debrief. A platform like Spidergap saves the form-building, the group averages and the charts, and leaves the comment theming to you, which is the step covered above. Whether AI tools built for coaches beat plain ChatGPT applies the same test to the rest of a coaching stack.
The parts of a 360 that stay human
- Names and identifying detail. They never go in, and the AI is told not to guess who wrote a comment.
- Personality or capability labels. A theme describes behaviour that raters reported. Asking an assistant what kind of leader the client is invites a verdict nobody can check.
- Deciding what matters. Which three themes go in the headline is your call, informed by the counts and the client's goals.
- Serious concerns. Allegations leave the report and follow the process in your agreement.
- The debrief itself. The conversation is the product, and a report that arrives without you reads as a verdict.
Questions coaches ask about AI and 360 feedback
How many raters does a 360 need?
Enough to protect anonymity. A common rule of thumb is at least three responses in a group before you show that group's scores. Most coaches invite eight to twelve people across manager, peers, direct reports and others, and expect a few not to respond. Fewer than three in a group means merging it into another group.
Should I tell raters that AI will read their comments?
Yes. Say so in the invitation: who reads the comments, that names are removed before anything goes into an AI tool, which kind of plan you use and when the data is deleted. Raters who learn about it afterwards lose trust in the process, and the client's report loses credibility with them.
Can AI write the whole 360 report?
It can draft summaries and theme descriptions, but the report is yours and carries your judgement. Treat AI output as a first pass to check against the raw comments: counts, verbatim quotes and the balance of strengths and development areas. Your name on the report means you stand behind every line.
What if a comment alleges serious misconduct?
Do not fold it into a theme. Set it aside and follow what your coaching agreement and professional body say about serious concerns, taking advice if you need it. A 360 report is developmental and an allegation needs a different process, so say that clearly at the contracting stage.
How long does a 360 take from start to debrief?
Allow four to five weeks: one for contracting and rater selection, two for collection with reminders, one for cleaning, theming and the report, then the debrief. Rushing collection lowers the response rate, and thin data makes every theme in the report weaker.
Further reads
- ChatGPT Prompts for Executive Coaching: Prep, Reflection and Goals — Prompts for session prep and goal setting around the debrief.
- AI Tools for a Coaching Business: Notes, Programmes and Follow-Up — Notes and follow-up tools across a whole coaching practice.
- How Business Coaches Use AI Without Losing the Personal Touch — Using AI without losing what clients hire you for.
- AI Coaching for Leaders: What to Offer Clients Alongside Your Sessions — What to offer clients alongside your human sessions.
- AI for Life Coaches: Between-Session Support Within the ICF Code — Between-session AI use inside a professional code.
- What to Check in an AI Tool's Privacy Policy and Terms — A way to read any AI tool's privacy terms yourself.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Spidergap pricing page, Your360.ai home and pricing pages, smart360 home page, Google Docs Editors Help pages on publishing forms, and the OpenAI Business pricing page, all read in November 2026.