Title and thumbnail an episode
Give the team a title and thumbnail they can ship. Base the promise on what the episode delivers. Make the opening support that promise.
Match the work to the request. For a title request, give titles with a short thumbnail note. For an intro check, give the opening fix. For test results, go straight to the testing reference. Do not make the user complete the whole workflow.
Start with what exists
Read SHOW-PROFILE.md in the show’s working folder if it exists. It’s a short notes file about the show (audience, look, past titles, and lessons) that the Podcasting skills share. Use its audience, style rules, past titles, and lessons. Do not ask for facts it already holds. If it is missing, use a short working note: audience, show promise, format, look, recent titles, and known results. Mark guesses. A profile is optional.
Find the episode source, guest’s relevant proof, and available images. Use the user’s material first.
| Input | Action |
|---|---|
| Episode link only | Open the page and seek a transcript. Inspect images or footage if tools allow. Say what you could access. If the talk is unavailable, ask for notes or a transcript and draft only a provisional angle from the description. |
| Transcript only | Find supported promises. Use exact lines as evidence. Give a shot brief and mark images as needed. Do not claim to have seen expressions, props, or the set. |
| Bio only or no episode material | Ask what the guest says that a viewer would care about. Offer a short brief to fill in. Do not invent claims from the guest’s job. |
| Existing drafts | Review those first. Keep the strong parts. Give a replacement for each material weakness. |
For a fresh show, ask who should watch and why. If a deadline is tight, state a working audience and proceed with a provisional brief.
Find the strongest promise
Read enough of the talk to catch its main point and its limits. Write a short evidence table: possible promise, source line or time, and what the claim leaves out.
Find the wider need inside the topic. A narrow craft lesson may help people think more clearly. Keep the wider promise within what the guest explains. Favor a lasting question when the episode supports one. For a news show, use the event and date that make the episode useful now.
Include a title led by the guest’s strongest relevant proof. Use a name first when the audience knows it. For a less known guest, test a clear role, result, or credential. Verify public credentials with a reliable source. Label user-supplied facts that still need a check. Never turn an old role into a current one.
If outside examples would resolve a weak angle, study a few comparable episodes. If the Benchmark MCP is connected, use it to find real outlier episodes, titles, and thumbnails. An outlier is an episode that did much better than that show’s usual. Otherwise use public pages or user examples. Compare similar formats and ages. Borrow the reason an idea works. Keep this show’s look.
Build pairs
Explore a few distinct angles before changing words within one angle. For a full brief, draft 6 to 10 pairs and narrow to up to three. For a quick request, give the best pair and one useful alternative.
For each shortlisted pair, include:
- Exact title and character count. Put the main reason to watch early. Keep within YouTube’s 100-character limit. Check the visible text on target screens; 60 characters is a useful review point, not a fixed cutoff.
- Exact thumbnail text, including a no-text option when the image works alone. Let the text add a clue or tension the title leaves open.
- Photo or frame ID, crop, face size, expression, text position, and color notes. Mark missing assets as a shot request.
- Evidence for the promise and the viewer it should attract.
Use real frames or shoot photos when they fit the show. Preserve the guest’s expression and the meaning of any object. Do not invent a prop as if it appeared on set. A designed image can work for an audio show; label it as artwork.
Match the look to the talk. A calm interview may need space and a thoughtful face. A comedy show may need visible play. Add the host when it helps people see that this is an interview. Keep the set’s useful visual cues.
Follow the show’s rules for type, color, and framing. Check the pair at phone size and beside recent episodes. Read the title and thumbnail together. Remove repeated words when they add no meaning. Keep enough contrast to read the text.
For a major launch or a well-funded test, build 15 to 20 thumbnail drafts within the chosen system. Vary the base shot, crop, and text with a clear reason for each group. These are design drafts. Send only the best options into a live test. Scale down when the team has few assets or little traffic.
Check the opening
Review the first 30 to 60 seconds as a starting window. Follow the actual opening structure if it runs longer. Do not assume it ends at the guest introduction.
State what a new viewer expects from each final pair. Find the first line or shot that meets that expectation. Check whether the opening starts to answer the question and gives a reason to trust the guest. It need not finish the full answer at once.
If the opening wanders, name the exact line to move, or write a short host pickup. Label new script as proposed wording. Keep source quotes exact. Preserve any caveat that changes the claim.
Check visual continuity when footage is available. YouTube can show previews before a click, but the preview is not always the intro. Do not promise control over its selected frames.
Use one verdict: Ready, Revise, or Cannot verify. Say which missing source prevents a stronger verdict.
Test when there is a real choice
Load references/testing.md when planning a live test, comparing tools, reading results, or deciding whether to test again. If you can’t open it, continue with the steps here. Skip it for a simple writing request.
Choose the next round from the uncertainty. Use a pair test to choose between whole promises. Hold the title fixed to learn about images. Hold the image fixed to learn about wording. Keep every tested promise true to the same episode.
Return a usable brief
Lead with the recommended pair and why it fits. Then give:
- Up to three final pairs, with source evidence and complete image briefs.
- The opening verdict and exact edit if needed.
- Missing assets or facts that affect readiness.
- One test question and next round, only if useful or requested.
Give rejected directions only when they explain a real tradeoff. Save the brief in the user’s episode folder when file work is requested. Keep a stable ID for each pair so a designer and test report can refer to the same option.
For an invented example, use Better Questions, a calm interview show for busy adults. Its fictional guest, sleep researcher Dr. Lena Park, discusses sleep tracking. A supported pair might be “Sleep Researcher: When Your Sleep Score Becomes a Problem” with “THE MORNING CHECK” over a real frame of her looking at a phone. Use that frame only if supplied. The opening should reach the sleep-score question before unrelated host updates.
If installed, use podcast-clips next to select moments from the same episode. Use podcast-growth-review if later results need a broader diagnosis. If neither is installed, the brief above still gives the team what it needs to publish and test.