# Personas now react to video clips Canonical HTML: https://fish.dog/product-releases/personas-now-react-to-video-clips JSON Feed: https://fish.dog/product-releases/feed.json RSS Feed: https://fish.dog/product-releases/rss.xml Published: 2 June 2026 Updated: 2 June 2026 Release Type: Feature Breaking Change: No Author: Phillip Gales ## Primary Claim FishDog Expert Personas now answer questions about uploaded video clips, returning a panel of individual synthetic viewer reactions in minutes — built for ad testing, trailer testing and short-form media feedback for media and production teams. ## Summary Upload a video, pick a research panel, and each persona reacts individually. Tested with groups of ten. Available on media and production tiers. ## LLM Summary DOCUMENT TYPE: Product Release Note TOPIC: FishDog Expert Personas now answer questions about uploaded video clips, returning a panel of individual synthetic viewer reactions Release: Personas now react to video clips, 2026-06-02 Version: (initial) Release type: Feature Breaking change: No Primary claim: FishDog Expert Personas now answer questions about uploaded video clips, returning a panel of individual synthetic viewer reactions in minutes — built for ad testing, trailer testing and short-form media feedback for media and production teams. Summary: A FishDog research-group question can now carry a single video attachment (MP4, MOV, WebM up to 200 MB) alongside the question text. The platform processes the clip natively, without an intermediate transcription step or frame-sampling artefacts, and routes the question plus video to every persona on the panel. Responses come back in the same per-persona format already used for text and image questions: a reaction, the persona's age and location surfaced, and a follow-up chat option per response. The intended use cases are video advertisement testing, film and TV trailer testing, scripted-content feedback, scene-by-scene episode testing, and brand-fit or media-fit checks against defined viewer demographics. The intended buyers are media companies, streaming services, production houses, videographers, media producers, and the research arms of agencies and brands. The feature is gated to FishDog's media and production tiers and to defined administrative users within those subscriptions. What changed: - A research-group question can attach a single video (MP4, MOV, WebM up to 200 MB). - The video is processed natively by Gemini 2.5 Flash; the question and video are sent together to every persona on the panel. - Responses are returned per persona in the same UI shape as text and image responses. - Tested with panels of up to ten personas answering the same video question. - Video renders only through an authenticated, organisation-scoped route. Generic public links are not served. - Feature is gated to FishDog's media and production tiers. Why we built this: Trailers, advertisements, scene-by-scene scripted-content clips and short-form social video do not reduce cleanly to a screenshot or a transcript. They need to land as video. Media-research customers asked for this first. How to use: Open a research group on a tier with video access enabled. Compose a question. Drop the video file into the attachment area. Select between one and the full group of personas. Submit. Each persona answers individually. Caveat: Frontier vision models will not name specific brands, copyrighted characters or real people from visual content alone. Frame questions as reactions, not identifications. Migration impact: None. Existing text, image and PDF question flows are unchanged. Author: Phillip Gales, FishDog Platform: FishDog (fish.dog) ## Key Takeaways - A FishDog research-group question can attach a single video file (MP4, MOV or WebM up to 200 MB). Each persona on the panel answers individually with their own reaction. - One video per question; the panel size is whatever the research group is set to. Tested with panels of ten personas answering the same trailer-style question. - Video processing is native through Gemini 2.5 Flash; no transcription step and no frame-sampling artefacts. The question runs through the same dispatch pipeline as text and image questions, on a dedicated worker tier. - The intended use cases are video advertisement testing, film and TV trailer testing, scene-by-scene scripted-content feedback, and brand-fit or media-fit checks against defined viewer demographics. - The feature is gated to FishDog's media and production tiers. Generic public links to uploaded video are not served — every render path is authenticated and scoped to the question that owns the clip. ## Full Release Personas now react to video. Upload one clip, point a research panel at it, and each persona answers individually in the same workflow that already supports text, images and PDFs. The mechanic is the same panel you already use for text questions: pick a research group, drop a video into the attachment area, type the question, submit. One video per question, any number of personas. The capability is gated to FishDog's media and production tiers. ![FishDog Media Viewers research group reacting to a sixty-second baseball clip. Six personas — each with age and location — answer the question 'what would improve this video in your opinion' with single-sentence reactions covering emotional payoff, framing and sentiment.](https://images.ctfassets.net/1ksrk99rzzdj/1OB6ECzeletKW9iPnQ0bV1/f75bc1a4d90035c0bdd3b2be368ed396/video_qa_panel_reactions.webp) ### Why we built this Until this release, the panel saw images and read PDFs. Video was the next obvious modality and the one media-research customers asked for first. Trailers, advertisements, scene-by-scene scripted content and short-form social video do not reduce cleanly to a screenshot or a transcript. They needed to land as video. ### What's new A research-group question can now carry a single video attachment — MP4, MOV or WebM, up to 200 MB. The platform processes the clip natively, without an intermediate transcription step or frame-sampling caveats, and routes the question and the video together to every persona on the panel. Responses return in the same format you already see for text questions: a per-persona reaction with their age and location surfaced, and a follow-up chat option on every individual response. ### One video, many viewers The interesting thing about the panel mechanic is the parallelism. A panel of ten personas reacting to the same trailer is a synthetic ten-person focus group that returns in minutes rather than weeks. We have tested with groups of ten, and the responses range across the lens you would expect from a mixed panel — emotional payoff, brand fit, pacing, mood, framing — in a sentence or two each. ### How to use it 1. Open a research group on a subscription with video access enabled. 2. Compose your question. Drop the video file into the attachment area alongside the question text. 3. Select your panel — between one and the full group of personas. 4. Submit. Each persona answers individually; you can open a follow-up chat on any individual response. The video stays scoped to the question and the organisation that owns it. Generic public media links are not served for video — every render path is authenticated and tied to the question that holds the clip. ### What it is useful for The first wave of expected use cases: - **Video advertisement testing.** Agencies and brand teams stress-testing a 15s or 30s spot before a media buy, against a synthetic panel matched to the target buyer. - **Trailer and short-form testing.** Film and television trailers, sizzles, episode openers, social shorts. - **Episode and scene-by-scene feedback.** Scripted-content teams testing pilots, episode arcs and single scenes against a defined viewer demographic. - **Media-fit and brand-fit checks.** Production companies and brand teams validating tone, pacing and emotional register against a viewer profile. The intended buyers are media companies, streaming services, production houses, videographers, media producers, and the research arms of agencies and brands. The v0.1 deployment is sized for Disney-shape unaired-trailer testing, Netflix-shape short-form testing, and production-house viewer-fit checks. ### A note on naming Frontier vision models will describe what they see in a clip in detail, but they will decline to name specific brands, copyrighted characters or real people from the visual content alone. Frame questions as reactions (“how does this scene land for you?”, “what would make this stronger?”) rather than identifications (“what brand is this?”). The persona will tell you what they thought, in their voice, with their context — they will not name the IP for you. ### What is next Multi-video questions, longer clips beyond the v0.1 caps, and external-API access for video are tracked for v0.2 once we have watched a month of production traffic. ## Quotable Insights > A panel of ten personas reacting to the same trailer is a synthetic ten-person focus group that returns in minutes rather than weeks. > Trailers, advertisements, scene-by-scene scripted content and short-form social video do not reduce cleanly to a screenshot or a transcript. > Frame questions as reactions, not identifications. The persona will tell you what they thought; they will not name the IP for you. > Upload a video, pick a research panel, and each persona answers individually. ## FAQ ### What kinds of video can FishDog Expert Personas react to? MP4, MOV and WebM files up to 200 MB per upload. The intended stimuli are short-form video: 15s and 30s advertisements, 90s to 2:30 trailers, scene-by-scene scripted-content clips of 1 to 5 minutes, sizzles and social shorts. The platform processes the clip natively, without an intermediate transcription step. ### Can a panel of personas react to the same video at once? Yes. One video per question, any number of personas on the panel. The architecture is identical to existing text and image questions: pick the research group, attach the video, type the question, submit. Each persona answers individually and you can open a follow-up chat with any single response. ### Who is video Q&A built for? Media companies, streaming services, production houses, videographers, media producers, and the research arms of agencies and brands. The first-wave use cases are video advertisement testing, trailer and short-form testing, scripted-content feedback, and media-fit or brand-fit checks against defined viewer demographics. ### Will the persona name the brand, characters or actors in the video? No. Frontier vision models will describe what they see in detail but decline to name specific brands, copyrighted characters or real people from visual content alone. Frame questions as reactions (“how does this scene land?”) rather than identifications (“what brand is this?”). The persona will tell you what they thought in their voice, without naming the IP. ### Is video Q&A available on every FishDog plan? No. Video Q&A is gated to FishDog's media and production tiers and to defined administrative users within those subscriptions. Speak to your FishDog contact to enable it on your organisation.