check_quotes
ChatGPTCheck which quotes can be used in a highlight reel, and how long the reel would run, before you call create_highlight_reel. Takes the same highlights list create_highlight_reel takes, and creates nothing. For each quote it returns found, and on a match matched_text (the words the recording holds, which can differ from the quote you passed), duration_seconds, and media_type (the recording a reel would play it from). A quote with found: false gives a reason and would fail create_highlight_reel, so reword it, drop its type, or drop the quote. duration_seconds at the top level is how long a reel of the found quotes would run — use it to pick quotes for a target length. It returns no file links: build the reel with create_highlight_reel and open the link it gives you.
create_asset_upload_url
ChatGPTGet a short-lived pre-signed URL to upload ONE media file (image/video/audio) directly to Voicepanel storage, for use as study stimulus. This is for agent/coding clients that can run a shell command (e.g. Claude Code); a plain chat client cannot use it — use create_upload_session (a hosted upload page) for that. Returns upload_url (a pre-signed PUT URL) and public_url. To upload, the client must run the following — WITH THE USER'S APPROVAL, since it reads a local file and runs a shell command: curl -X PUT "<upload_url>" -H "Content-Type: <mime_type>" --data-binary @<local-file-path> After the PUT succeeds, use public_url as the stimulus url in create_study/edit_study. Handles large files (e.g. video) that inline uploads can't. Accepted types: JPEG, PNG, GIF, WEBP, MP4, MOV, MP3, M4A, WAV, WEBM (SVG and other types are rejected). HTML is accepted only for accounts with the HTML stimulus feature enabled, and is rejected otherwise.
create_eval
ChatGPTCreate an eval that pairs a scenario set with a grader. You can leave out scenario_set_id and grader_id and attach them in the app later, but start_eval_run needs both. treatment_labels names the treatments to compare; each run adds its label to this list.
create_grader
ChatGPTCreate a grader: an LLM judge that scores each treatment on its rubrics. Give it at least one rubric. Pass rubric_ids to reuse rubrics from list_rubrics as they are, and rubrics to create new ones with the grader. Leave out system_prompt and user_prompt_template to use the default judge prompts. A custom user prompt template can use {{scenario_title}}, {{scenario_need}}, {{scenario_context}}, {{scenario_metadata}}, {{rubrics}}, and {{treatments}}.
create_highlight_reel
ChatGPTCreate a highlight reel of participant quotes and get a Voicepanel link to watch, edit, and export it as a video. Give each quote with its study_id and response_number, in the order the reel should play. The response number is what the dashboard shows, and is seq_num in the output of list_responses and get_response. Copy each quote from the transcript (get_response, get_analysis_question, or get_analysis_insight). Voicepanel finds it in the recordings and adds the timings; small edits to a quote still match. type requires one recording: audio, video, screenshare, or mobile_screenshare. A quote that recording does not hold fails, rather than falling back to another recording. Omit type to use whichever recording holds the quote. If any quote is not found in the recordings, nothing is created and the error lists each quote that failed, so you can fix those quotes and call again. On success it returns the reel id, its page URL, the number of highlights, and the reel's length. Use check_quotes to test the list first, and to see how long the reel would run.
create_intercept_distribution
ChatGPTCreate an intercept distribution — invites visitors of the user's own web application via the embedded Voicepanel snippet (shown as 'Intercept' in the Voicepanel app; stored as type 'website'). To qualify visitors before the study, pass screener_questions directly, or pass an audience description to have the screener generated automatically. Only use this when the user has chosen an on-site intercept; if they haven't said how they want to collect responses, ask them to choose between a link (create_link_distribution), an intercept, or panel recruiting (create_panel_distribution). Returns url, a dashboard link to view the distribution — share it with the user.
create_link_distribution
ChatGPTCreate a link distribution — a shareable link the user sends directly to their own respondents. To qualify respondents before the study (useful for reaching a niche audience by sharing the link widely and screening), pass screener_questions directly, or pass an audience description to have the screener generated automatically. Only use this when the user has chosen link sharing; if they haven't said how they want to collect responses, ask them to choose between a link, an on-site intercept (create_intercept_distribution), or panel recruiting (create_panel_distribution). Returns share_url, the public link respondents open to take the study, and url, a dashboard link to view the distribution. Note the distribution starts in draft — publish_distribution makes the share_url live.
create_panel_distribution
ChatGPTCreate a panel distribution — participants are recruited automatically from a research panel. Created in draft status only; this tool does not contact the panel provider. Publishing a panel recruit is not available over MCP — tell the user to publish it from the Voicepanel dashboard, which is where credits are spent and recruiting starts. Only use this when the user has chosen panel recruiting; if they haven't said how they want to collect responses, ask them to choose between a link (create_link_distribution), an on-site intercept (create_intercept_distribution), or a panel. Call suggest_panel_provider first to pick the provider. Pass an audience description to have targeting filters and a screener generated automatically (required for 'respondent'). Returns url, a dashboard link to view the distribution — share it with the user.
create_rubric
ChatGPTCreate a rubric: one numeric scale that a grader scores treatments on. Give each level a label and a description of what a treatment at that level looks like. If you leave out scale, the rubric gets a 0–5 scale from "Does not meet" to "Fully meets". To create rubrics and a grader in one step, pass the rubrics to create_grader instead.
create_scenario_set
ChatGPTCreate a scenario set from a list of user needs. Each scenario needs a need: one user need, in the user's words, e.g. "Find a waterproof jacket under $150 for a hiking trip". title is optional; the server derives it from the need when you leave it out. metadata holds string attributes you can group results by later, e.g. { "segment": "new_user" }. Send at most 200 scenarios per call. For a larger list, split it into several sets.
create_segmentation
ChatGPTCreate a new segmentation to categorize responses for a study
create_study
ChatGPTCreate a new study. If you are writing the plan yourself, call get_study_design_guide first — it covers the plan fields, the research rules this account applies, and the constraints that fail validation. Returns url, a dashboard link to open the study — share it with the user.
create_treatments
ChatGPTAdd one treatment label to the scenarios of a scenario set. A treatment is what one variant shows for a scenario, e.g. the search results of the current ranking ("control") or of a new one ("variant-b"). An eval run grades one label across the scenarios that have it. Give each treatment the scenario_id from get_scenario_set and an artifact. The judge reads only the artifact's summary, listings, and pages, so each artifact needs at least one of them. Use listings for structured result rows. A scenario holds one treatment per label: if a scenario already has this label, the call fails unless you pass replace: true. With replace, only the scenarios in this call change. Send at most 100 treatments per call; call again with the same label for more.
create_upload_session
ChatGPTCreate a short-lived web link the user can open to upload their own media (images, video, or audio) from their device for use as study stimulus. Returns upload_url — present this link to the user and ask them to open it and upload their files. Once they confirm they are done, call get_upload_session with the returned session_id to retrieve the uploaded files' public URLs, then use those URLs as stimulus in create_study/edit_study. Use this when the user has a local media file to add; media that is already hosted at a public URL (a website, a YouTube link, a Figma prototype, etc.) can be referenced directly in the stimulus without uploading.
edit_distribution
ChatGPTEdit an existing distribution. Panel distributions can only be edited while in draft status.
edit_study
ChatGPTEdit an existing study. If you are writing the plan yourself, call get_study_design_guide first — it covers the plan fields, the research rules this account applies, and the constraints that fail validation.
generate_study
ChatGPTGenerate a complete study using AI from a natural language description. The AI will create a full study plan with sections, questions, and conversation guides based on best practices in user research. Use this when you have a research objective or description and want the AI to design the study structure. Use create_study instead if you want to manually specify study fields. Returns the study with url, a dashboard link to open it — share it with the user.
get_account
ChatGPTGet current account information (name, subscription tier, credits, limits)
get_analysis_insight
ChatGPTGet a full analysis insight with evidence quotes and response IDs
get_analysis_insights
ChatGPTGet AI-generated analysis insights (the findings/summary) for a study — the latest completed summary, evidence omitted. Start here for high-level findings. Use get_analysis_insight for full evidence quotes. The returned _versions array lists prior summary versions you can fetch by id. Pass analyze_in_original_language to view the summary in the study's original language instead of English.
get_analysis_question
ChatGPTGet a full analysis question with evidence quotes and response IDs per category
get_analysis_questions
ChatGPTGet analysis questions with summaries and answer categories for a study — the latest result per question (continuously maintained, not tied to a run), evidence omitted. Use get_analysis_question for full evidence quotes. Pass analyze_in_original_language to view summaries in the study's original language instead of English.
get_distribution
ChatGPTGet detailed distribution information including config, incentive, quota, and targeting. Returns url, a dashboard link to view the distribution, and — for link distributions — share_url, the public link participants open to take the study (share this one with respondents).
get_eval
ChatGPTGet one eval and all of its runs, oldest first. Each run gives its treatment label, status, and summary: mean_scores per treatment on a 0–1 scale, and the number of scenarios graded. Use get_eval_run for the score and rationale of each scenario.
get_eval_run
ChatGPTGet one eval run: its status and summary, and one result per graded scenario. Each result gives the scenario title, overall_score on a 0–1 scale, and scores keyed by rubric ID, with the raw scale_value and the judge's rationale. While status is "running", graded_count rises toward scenario_total.
get_highlight_reel
ChatGPTGet a highlight reel: its name, its page URL, and each highlight in the order the reel plays them. Each highlight gives its index, study_id, response_number, the quote it shows, its duration_seconds, and trimmed (true when a person trimmed the clip in the editor). Call this before update_highlight_reel: pass { keep: index } to keep a highlight as it is.
get_response
ChatGPTGet a single response with full conversation transcript
get_scenario_set
ChatGPTGet one scenario set with its scenarios in order, and the treatment labels attached to it. A treatment is the artifact one variant shows for a scenario; treatment_labels gives each label and how many scenarios have it. Pass a label to start_eval_run. Add treatments with create_treatments.
get_screener_design_guide
ChatGPTHow to write screener questions that qualify the right respondents and resist gaming. Covers the qualification tags and the must_select/must_one_of trap, obfuscating the target so respondents can't guess the qualifying answer, honeypot options, option randomization and anchoring, early screenout, question order, and how to write open-ended and matrix questions. Includes this account's own AI rules if it has customized them. Call this before passing screener_questions to a distribution tool by hand — passing an audience description instead applies this guidance automatically. Read-only.
get_segmentation
ChatGPTGet full segmentation details including segment definitions and matching criteria
get_stimulus_metadata
ChatGPTRead the caller-defined attributes stored on a stimulus — the tags used to group and compare stimuli during analysis (e.g. which model or price tier a variant represents). Stimulus IDs come from the plan returned by get_study, where each stimulus carries its id and, once set, its metadata.
get_study
ChatGPTGet detailed study information including questions, plan, settings, and response statistics, plus url, a dashboard link to open the study.
get_study_design_guide
ChatGPTHow to design a good Voicepanel study, and how to express it as a study plan. Covers research best practices (including this account's own AI rules, if it has customized them), every step and stimulus type with its exact fields, the template variables you can interpolate into question text and stimulus URLs, conditional sections, section randomization, and the constraints that fail validation. Call this before writing or editing a plan by hand — generate_study already applies this guidance internally. Read-only.
get_study_stats
ChatGPTGet response statistics for a study (counts by status and distribution).
get_upload_session
ChatGPTRetrieve the media a user uploaded via an upload session (see create_upload_session). Returns the uploaded files as { url, stimulus_type } — use each url as a stimulus in a study plan (create_study/edit_study). Call this after the user confirms they have finished uploading; status stays 'awaiting_upload' until at least one file has been uploaded.
get_user_profile
ChatGPTGet current user profile (name, email, role in account)
list_distributions
ChatGPTList all distributions for a study
list_evals
ChatGPTList the evals in this account, newest first. An eval pairs a scenario set with a grader. Each run of an eval grades one treatment label across the scenarios. Each eval carries its run count, its three newest runs, and score stats per treatment. Use get_eval for every run.
list_graders
ChatGPTList the graders in this account, newest first. A grader is an LLM judge: a model, a system prompt, a user prompt template, and the rubrics it scores on (rubric_ids).
list_responses
ChatGPTList responses for a study (excludes full conversation transcripts — use get_response for that)
list_rubrics
ChatGPTList the rubrics in this account, newest first. A rubric is one numeric scale, from 0 upward, with a label and description for each level. A grader scores each treatment on one or more rubrics.
list_scenario_sets
ChatGPTList the scenario sets in this account, newest first, with the number of scenarios in each. A scenario set is a list of user needs that an eval grades treatments against.
list_segmentations
ChatGPTList segmentations for a study (name, key, description). Use get_segmentation for full segment definitions.
list_studies
ChatGPTList the active studies in the account, with response counts and a dashboard url per study. internal_name is the study name as it appears in the dashboard. Archived studies are excluded.
list_tags
ChatGPTList all labels in the account with the number of studies using each. Labels are stored as tags; a study has a single label.
list_templates
ChatGPTList the templates saved in the account. Each row carries a type of 'distribution', 'screener', or 'typing_tool'. A 'typing_tool' row is the account's typing tool, not a template to copy: a researcher uses it on a study from the study editor.
publish_distribution
ChatGPTPublish a distribution to make it live and start collecting responses. For link distributions, returns share_url — the public link to share with respondents, now live. Supports link and intercept distributions; publishing panel distributions is not currently available over MCP (publish those from the Voicepanel dashboard).
review_study
ChatGPTGet an AI review of a study's configuration before publishing. Returns a score (1-5) and improvement suggestions. Results are cached per study version.
run_analysis
ChatGPTGenerate a fresh analysis summary (findings/insights) for a study from its responses — the same action as the Update button on the analyze page. This does NOT regenerate per-question analysis (that is maintained automatically as responses complete); it produces a new summary version asynchronously. Requires at least one analysis question to have finished processing.
set_stimulus_metadata
ChatGPTTag a stimulus with caller-defined attributes so analysis can group and compare by them later. Pass metadata to set several attributes at once, or key plus value to set a single one — exactly one of the two. By default metadata is merged into what is already there; pass replace: true to drop every attribute it doesn't name. To remove one attribute, pass its key with a null value; to clear them all, pass an empty metadata with replace: true. Stimulus IDs come from the plan returned by get_study.
start_eval_run
ChatGPTStart a run that grades one treatment label across the eval's scenarios. The grader runs in the background and the tool returns at once with the run status "running". Call get_eval_run with the run id to see progress and scores. The first run freezes the eval: after it, the eval's scenario set and grader cannot change. Get the treatment labels from get_scenario_set. scenario_limit grades only the first N scenarios, for a quick check.
suggest_panel_provider
ChatGPTGiven a plain-text description of the target audience, recommend which research panel to recruit from, with a rationale and ranked alternatives. Call this before create_panel_distribution to decide the provider — then present the recommendation to the user and let them override. Returns recommended_provider (the best-fit panel this account can use — 'user_interviews' or 'respondent'), rationale, best_fit_label (the ideal panel, which may be a specialty we don't self-serve), requires_contact (true when the ideal fit isn't self-serve), and options (accessible panels ranked best-fit first).
update_highlight_reel
ChatGPTRename a highlight reel, or replace its list of highlights. highlights replaces the whole list, in play order. A highlight you leave out is removed from the reel. To keep an existing highlight, with any trim a person made in the editor, use { keep: index }. index is the position of the highlight in the list get_highlight_reel returned, not a response number. To add a highlight, use a quote item in the form create_highlight_reel takes. Indexes change after each update, so call get_highlight_reel before every update. Example: to add a quote to the end of a reel of 3 highlights, pass [{ keep: 0 }, { keep: 1 }, { keep: 2 }, { study_id, response_number, quote }]. Omit highlights to change only the name. If any quote is not found in the recordings, or a keep index is not in the reel, nothing is changed and the error lists each item that failed. Use check_quotes to test new quotes first. On success it returns the reel id, its page URL, the number of highlights, and the reel's length.
update_response
ChatGPTUpdate a response's status and/or whether it is included in analysis
create_asset_upload_url
Claudecreate_intercept_distribution
Claudecreate_link_distribution
Claudecreate_panel_distribution
Claudecreate_segmentation
Claudecreate_study
Claudecreate_upload_session
Claudeedit_distribution
Claudeedit_study
Claudegenerate_study
Claudeget_account
Claudeget_analysis_insight
Claudeget_analysis_insights
Claudeget_analysis_question
Claudeget_analysis_questions
Claudeget_distribution
Claudeget_response
Claudeget_segmentation
Claudeget_study
Claudeget_study_stats
Claudeget_upload_session
Claudeget_user_profile
Claudelist_distributions
Claudelist_responses
Claudelist_segmentations
Claudelist_studies
Claudelist_tags
Claudelist_templates
Claudepublish_distribution
Claudereview_study
Clauderun_analysis
Claudesuggest_panel_provider
Claudeupdate_response
Claude