Find content ideas with Claude

One connector and one prompt. It reads Reddit threads and 1–2 star reviews across your niche, then hands back a scored spreadsheet of what people actually complain about.

Step 1: Create a free Apify account

The free plan gives you $5 in credits every month. A full pass costs roughly 30 cents to a dollar, so that's about 5–10 runs a month. Plenty, since you should run this weekly at most. Credits reset monthly and don't roll over. If you run out, Apify stops running jobs. It doesn't bill you.

Step 2: Connect Apify to Claude

Open Claude → Settings → Connectors
Find Apify in the directory
Click Connect and authorize with your new account

While you're there, connect Google Drive too. You'll use it in step 4.

Step 3: Paste the prompt

Start a new chat, copy the prompt below, fill in the blank lines and send it.

NICHE: be specific. "AI tools" is too broad. "AI writing tools for marketing agencies" gives real complaints.
AUDIENCE: who exactly. Role, stage, situation.

Then leave it alone. It won't check in until it's done. A quick pass takes a few minutes.

Step 4: Open the CSV in Google Sheets

You'll get back 20–25 scored ideas. In the same chat, ask:

Save this CSV to my Google Drive as a Google Sheet.

Claude writes it to your Drive and gives you the link.

What comes back

Column	Meaning
idea	The problem in one sentence
score	Out of 9: recurrence + heat + specificity
audience	Who has this problem, and at what stage
their_words	How they describe it, in their language
hook_line	An opening line you can film
source_url	Where it came from
status	unused / drafted / published (you update this)

Three things that will save you a run

Don't run it daily. Pain points don't change every 24 hours. Run it weekly and draw from the sheet each day.
Check the top scores. If your top five are all 9/9, the scoring stopped working. Ask it to re-score more strictly.
Sanity-check the winner. Pulling only 1–2 star reviews means billing complaints show up everywhere by design. That's your filter, not a discovery.

Read the their_words column first. It's the vocabulary your audience actually uses, and the one thing you can't recreate without re-running the research. Your hooks come from there.

The prompt

Fill in the blank lines. Leave the rest as it is.

Prompt

Run a content research pass and build me a scored idea backlog.

NICHE:
AUDIENCE:
DEPTH:

## Sources - work down the list, don't stop at one
Use Apify with the named Actors below. Do NOT use apify/rag-web-browser for Reddit or YouTube - it routes through Google and gets blocked. If Apify isn't connected, use web search and tell me the pass will be shallower.

1. 1-2 star reviews of tools, courses and products in my niche (Trustpilot, G2, Gumroad, App Store). Start here - negative reviews are specific and describe a workflow that actually broke.
2. Reddit -> Actor: trudax/reddit-scraper-lite
   Reddit's own relevance ranking returns junk for phrase queries. Instead of passing my queries into `searches`, build direct subreddit search URLs for the 5-6 subreddits where my audience actually posts, and pass those as startUrls. Set skipComments=false - the comments carry the real language, not the post.
3. YouTube comments on top videos in the niche
   -> Actor: streamers/youtube-comments-scraper
4. Quora and niche forums

If an Actor errors, say which one and what the error was before falling back. Don't silently switch to web search.

## Search by sentence shape, not keyword
Keywords alone return generic discussion. Pair my topic with each of these and run every combination as its own query:
"I hate" - "I wish" - "does anyone else" - "why is there no" - "sick of" - "wasting hours on" - "better alternative to" - "gave up on" - "unpopular opinion" - "PSA:" - "underrated"

## Budget
quick = 6 searches / 15 items - standard = 12 / 30 - deep = 20 / 60
80% evergreen (top, past month), 20% fresh (past week).
Stop early if three searches in a row surface no new theme, and tell me you did.

## Score each out of 9
Recurrence (1-3) + Heat (1-3) + Specificity (1-3). Keep 5+, drop the rest.
Reserve 9/9 for ideas that genuinely max out all three. If everything at the top scores 9, the scoring isn't doing its job.

## Rules
- Read-only. Don't post, vote, or interact with anything.
- Skip self-promo disguised as a question ("I built X, thoughts?").
- No usernames or profile links. Link the thread, not the person.
- Rewrite any phrasing in your own words.
- Don't pad. If the niche is thin, say so and show me what you found.
- Flag anything that's an artifact of how I filtered rather than a real finding. If I only pulled 1-2 star reviews, billing complaints will be over-represented by design - say so instead of calling it a discovery.
- Run the whole pass without checking in. Surface once, at the end.

## Output
A CSV of 20-25 rows with these columns:
id, date_added, idea, score, recurrence, heat, specificity, times_seen, audience, tone, their_words, hook_line, source_url, status

their_words = 2-3 phrases in the audience's real vocabulary, rewritten in your own words. This is the most important column - it's the only thing that can't be regenerated later without redoing the research. Never leave it thin.
status = unused for new rows.

Then, briefly: the top 3 with one line each, which sources actually worked and which failed, and which single row I should build first.