A spoken method for turning feedback notes into decision-ready themes.

Voice Prompting for Product feedback analysis works best when the analysis is tied to a decision. Start the prompt by naming the decision the work will inform: whether to rebuild onboarding, which of three signup pages to test, or what to fix before the next release. Then name the product area, the time window, and the customer segment the notes come from. Without that frame, a theme list becomes an interesting document that changes nothing. Say who will read the analysis and what they will do next. A theme that cannot be linked to a decision is trivia, no matter how often it appears.
Paste the feedback notes directly into the prompt and describe how they were gathered: support conversations, interview transcripts, survey comments, or sales call summaries. Say how many items there are, roughly how long each is, and whether they are verbatim or paraphrased. Note any known gap, such as missing enterprise customers or a week of unanswered tickets. Ask the assistant to work only from the pasted text and to say when the notes do not support a theme. This instruction prevents the response from importing assumptions from the wider market and presenting them as findings. Write the item count in the first line so the reader knows the scale.
Define what counts as one item before asking for themes. Is it a conversation, a comment, an account, or a complaint? A consistent unit makes counts meaningful and stops one long thread from dominating the result. Ask for the unit to be stated at the top of the response, along with the number of items examined. If the notes mix units, ask for them to be separated first. When a theme depends on two items that may be the same customer, ask for the overlap to be flagged. Counting is only useful when the reader knows exactly what was counted.
Write the acceptance criteria before you dictate. The analysis should contain a short answer to the decision question, a theme list with counts, two representative quotes per theme, a severity or frequency note, and a list of what the notes cannot answer. Each theme must be traceable to specific items. Ask for themes that contradict each other to be kept separate rather than merged into a balanced summary. If the response cannot meet these criteria, it should say so instead of filling space. These criteria turn a pleasant summary into something a product group can argue with. Number the criteria so a reviewer can refer to them by number.
Draft a provisional theme map before the prompt, then invite the assistant to revise it. Start with the categories your team already uses: onboarding, reliability, pricing clarity, integrations, performance, and support. Add two open categories, other and unclear, so unusual feedback has somewhere to go. Say that the map is provisional and that new themes are welcome if at least two items support them. Ask for a list of proposed changes to the map before the detailed analysis. Reviewing the map first prevents a long response built on categories that do not fit the product. Keep the map to a single page so it remains a working tool.
Choose granularity deliberately. Themes that are too broad, such as usability, hide the specific problem and produce advice nobody can act on. Themes that are too narrow produce a list of one-off complaints. Aim for themes a product manager could assign to a team and a release. Test the map by asking where three real quotes would land; if two quotes fit three themes, the map needs tightening. Ask for a parent theme and a child detail, such as export limits under reporting. Write the granularity rule in the prompt so the response does not drift between levels. Name one example under each parent so the intended detail level stays visible.
Decide which cuts matter before the analysis starts: plan tier, company size, tenure, region, or role. Ask for each theme to show whether it is concentrated in one segment or spread across all of them. Warn the assistant that a small segment can produce a loud theme, and ask for the item count behind each cut. If a cut has too few items to support a conclusion, ask for it to be labeled as a signal rather than a finding. These labels keep the analysis honest when someone later asks whether large accounts really care about the issue. Ask for the count and the share side by side rather than only a percentage.
Separate frequency from severity in the prompt. Frequency is how many items mention the theme; severity is how much the problem blocks the customer. Ask for both, with a short reason for each severity rating. A rare outage and a common wording annoyance should not land in the same priority column. Ask the response to name the evidence for severity, such as a canceled renewal or a described workaround. When severity is inferred rather than stated, require the label inferred. This distinction is what product groups use to decide what to fix first. Put the two measures in separate columns so nobody conflates them during discussion.
Ask for quotes exactly as they appear in the pasted notes, with a short identifier rather than a name. Say that paraphrases must be marked as paraphrases. When you dictate the instruction, be explicit: preserve the customer phrasing, including errors, and do not tidy grammar. Tidied quotes lose the signal that tells a team how the customer actually describes the problem. Ask for at least two quotes per theme and no more than four, so the evidence is representative without turning the response into a transcript. If a theme has only one quote, ask for it to be labeled thin.
Ban impression words from the theme statements: users love, customers hate, everyone wants. Ask instead for counts and quote evidence. A theme statement should read like a claim with support, such as seven of forty items mention an export limit, with two quotes. When the notes contain strong emotion, report the emotion as part of the quote rather than as the analyst conclusion. This rule keeps the analysis defensible in a review where someone asks how you know. It also prevents a vivid complaint from being mistaken for a pattern. Write theme statements in the same grammatical shape so they are easy to compare.
Contradictions are normal in feedback and should be preserved. Ask for a section that lists direct conflicts, such as requests for more automation next to requests for more control. For each conflict, ask for the item counts and the segments where each view appears. Do not let the response average the two views into a balanced statement that satisfies nobody. Then ask which conflict needs a decision from the product group and which can be resolved with better wording. A good analysis makes the conflict easy to discuss instead of hiding it in a smooth summary. Keep conflicts visible in the deliverable rather than resolving them in a private note.
Dictating the quote-handling rules is faster than typing them and easier to keep complete. Say the rules in the order you will check them: identifier, verbatim rule, paraphrase label, minimum and maximum quotes, thin-theme label. Voiskey removes filler words, stutters, and self-corrections, so the rule list arrives as clean instructions rather than a transcript of your thinking. Read the list back once and cut any rule you will not enforce. A short list you actually apply produces more consistent analysis than a long policy nobody remembers. Store the list beside the brief and read it once before each new analysis.
Voiskey keeps the review inside the document where the analysis lives. Open the notes beside the draft, press the shortcut, and speak your check for each theme: show me the items behind this count, and the two quotes I asked for. The polished request lands at the insertion point, so the review trail stays with the analysis. Read the theme statement aloud and ask whether a colleague could reproduce the count from the notes. If the answer is no, the theme is not ready. The workflow of speak, polish, type, and edit makes this check quick enough to run on every theme.
Recount at least three themes by hand. Pick one large theme, one small theme, and one that surprised you. Compare your count with the response and note any difference. Common causes are duplicate items, items that mention the topic without complaining, and quotes that were counted twice. Record the cause next to the theme so the next analysis avoids it. If the counts differ by more than a small margin, ask for the full item list rather than a summary. Counting errors undermine trust faster than any wording problem, because they are easy for a reader to check. Write the recount number next to each checked theme so the next reviewer sees it.
Return to the decision question at the top of the prompt. Ask whether the analysis actually answers it, or merely describes the themes. A useful test is to read the first paragraph aloud to someone who does not know the notes and ask what they would do next. If they cannot answer, the recommendation is buried or missing. Ask for a one-sentence recommendation, the strongest two themes behind it, and the main uncertainty. Keep that paragraph at the top of the deliverable. Most readers stop after the first paragraph and the theme titles; make both carry the argument. If the first paragraph hedges, rewrite it until the recommendation is plain.
End the review with a list of what the notes cannot answer. Typical items are why a behavior happens, how many silent customers share the view, and whether a workaround exists. For each open question, name the cheapest way to answer it: five interviews, a support tag review, or a short survey. Ask for the list to be ordered by the decision it would unblock. This list is often more valuable than the themes, because it directs the next round of research instead of inviting the team to over-read the current notes. Attach it to the analysis before sharing.
Turn the prompt into a repeatable brief with fixed blocks: decision question, notes description, unit of analysis, theme map, quote rules, segment cuts, and output shape. Save the blocks in the order you speak them, because the order is part of the method. Keep the decision question and output shape as the two blocks you edit every time; leave the rest stable. Test the brief on an old set of notes with a known conclusion and compare. A repeatable brief lets a new analyst produce a comparable analysis without relearning your standards. Print the blocks in order and keep the list beside the desk during a live analysis.
Share the brief with the product and research groups and agree on the theme map together. The team plan offers shared vocabulary and workflows, so category names and quote rules stay aligned when several people run analyses. Keep one canonical copy and a dated change note. When two analyses disagree, compare the briefs before comparing the conclusions; differences usually trace to a changed theme definition or a different unit of analysis. This practice turns a recurring argument about findings into a short check of method and evidence. Keep a short version log with dates and the person who changed each block.
Feedback often arrives in more than one language. Voiskey supports more than one hundred languages and translation, so an analyst can speak the brief in one language and ask for themes in another, or ask for quotes to be translated with the original kept beside them. Keep identifiers stable across languages and review translated quotes with someone fluent. Ask for a note when a translation cannot preserve a nuance. Multilingual analysis is only useful when the reader can see which words are the customer's own and which belong to the translator. Label the original and the translation in the same order for every quote.
Download Voiskey for Mac and Windows and run one real feedback set through the full brief before you standardize it. Choose a decision the team is actually facing, paste the notes, speak the blocks in order, and check three counts by hand. Keep the version of the brief that produced a defensible analysis and add the date. Then run it again on the next set and compare the output shape, not just the findings. Two runs are enough to see whether the brief is stable, and the comparison is the evidence your team needs to trust the method. Archive both runs with the brief version so a later analyst can repeat the comparison.
Voiskey is an AI voice typing app. Press a shortcut, speak the analysis brief, and the text appears at the insertion point in the app you are using. It removes filler words, stutters, and self-corrections and turns the spoken request into an AI-ready prompt, so the decision question, theme rules, and quote rules arrive clean.
Yes. Voiskey learns names, jargon, acronyms, and spelling preferences, so feature names and internal terms stay consistent across a long analysis brief. Say them once as you dictate and reuse the same brief. The app polishes the request and places it at your insertion point with your preferred spellings.
Yes. Voiskey turns spoken requests into AI-ready prompts, including file names, paths, and commands. For feedback work you can say the decision question, the theme map, and the quote rules in one pass, then edit the result. Unclear phrasing, repetition, and false starts are removed before the text appears.
Voiskey supports more than one hundred languages and translation. You can dictate the analysis brief in one language, ask for themes in another, and keep translated quotes beside the originals for review. A fluent reviewer should still check nuance, but the spoken brief and the analysis need not share a language.

Download Voiskey for Mac and Windows and dictate your next feedback analysis brief.
Use it across macOS, Windows, iOS, and Android