Parrot AI topic selection: from one signal to a shoot-ready idea pool

›

Audience comments312I can never find that prompt again

DMs and questions48Anything that works for weekly reports?

FAQs17The same question, asked three times over

On-site searches26AI weekly report template

Benchmark gaps4Non-technical roles × written tests

Primary sources9Model launch · tool price changes

Last cycle's review6The workflow post had the best completion

Comments, DMs and searches are illustrative · the structure and gate decisions come from the pipeline itself

Ideas aren't invented.
They're picked up from your audience.

Brainstorming in isolation produces “I think”. The real signals are already in your comments, DMs and search terms — the agent gathers them, deduplicates, traces each one back, and hands them to the next step.

Comments and DMs312 + 48

A question asked three times is an idea. Repeated confusion is more reliable than any burst of inspiration.

I can never find that prompt again

FAQs and searches17 + 26

Write the title in the words people already search with. A search term is demand that already happened.

AI weekly report template

Gaps in the benchmark4

A topic, format, audience or stance nobody holds is differentiated by default. Run a benchmark first →

non-technical roles × written tests

Primary sources and last review9 + 6

Official notices, launches and price changes supply timing; last cycle's best completion supplies in-house evidence.

the workflow post had the best completion

Every signal must trace back to its source: a primary document, a verified quotation, a source log. A “fact” with no provenance never reaches an idea card.

Not every trend is yours.
Three gates first.

The usual mistake is treating “related” as “go”. The gates are independent; failing any one records a skip with its reason — no forcing it, and no pretending you did not see it.

FIT Fit

Does this belong to your IP's main line? Traffic that has nothing to do with your audience is not your trend.

Fails → skip: off-positioning

SAFETY Safety

Is the information public and citable? Could the wording break platform rules, privacy or repeat an unverified rumour? Prices and specs must trace to the official text.

Fails → skip: unverifiable or risky

TIMING Timing

Is the window still open? Once the launch is saturated, you are repeating what everyone has said — unless you bring a new angle or new evidence.

Fails → skip: window closed, no new angle

Every gate decision and every skip reason stays on record, so the next signal of the same kind does not make you think it through again.

One idea card.
Five things, none optional.

An “idea” with only a title is a wish. A shoot-ready idea states its angle, evidence, format and hook — and ends with a score you can compare.

#08topic · AI tools

I let 3 AI agents do my job for 5 days

Angle: not a tool review — a record of one real job replacement.

92/ 100

Angle~~Where you cut into the same subject. It decides how the piece differs from everyone else's — not the rhetoric of the title.~~

Evidence~~A comment, a DM, a search term, a benchmark gap or the official text — named, so it can be clicked back to.~~

Format~~Voice, lists, long form, or one idea split across platforms. Format is decided with the angle, not patched on later.~~

Hook~~The human sentence in the first three seconds. If you cannot write it, the angle is not specific enough.~~

Score~~Positioning, audience match, domain knowledge, trend, benchmark gap and hit signals — for ordering the pool, not for gut feelings.~~

Once you approve it,
the ID follows it everywhere.

The pool has three states: awaiting, approved, backlog. The ID is not decoration — the schedule, the draft folder, the publish record and the review all key off it, so “whatever happened to that one?” always has an answer.

IDIdeaScoreStatus

#08I let 3 AI agents do my job for 5 days92Approved

#07Stop writing “You are a…”88Awaiting

#06After the price hikes, I kept 2 tools86Awaiting

#05Can AI actually own the weekly report?81Backlog

Test cheaply first

An uncertain idea does not need ten posts. Publish one 30-second test; if the data is flat, change the angle instead of pushing harder.

Skips are kept too

Signals stopped by a gate stay on record with their reason. Next time the same trend appears, you know why you passed.

One idea,
three platforms, three scripts.

After you approve an idea, the same angle is rewritten for each platform: TikTok needs a hook, RED needs a cover and a list, WeChat needs structure. They share the evidence, not the copy.

TikTok / DouyinConclusion first, process after~~Say the counter-intuitive thing in three seconds, then hold it up with the recorded test.~~

Hook: I did my job for five days. Then the AI did.

REDThe cover is the conclusion~~One point per page, with a copyable prompt or checklist.~~

Cover: 3 AIs did my job. I got 11 hours back.

WeChatLay the evidence out~~Process, failures, comparison, conclusion — for the people who read to the end.~~

Structure: why I tried → how → which step failed

Want to know who has done it and where nobody has? Parrot AI account benchmarking: turn the accounts that win into a map →

Your inbox isn’t the whole signal.
Half of it is in the data service.

Comments, DMs and FAQs are yours. Search terms, off-platform discussion, platform trends and benchmark gaps have to be fetched. The data service puts them in one place — in the order this page runs: size the demand, watch the platforms, check the facts, see the competition.

01 Size the demand~~Search volume, SERP and hot search — confirm someone is asking~~