I run an Etsy shop called Hajidraft, selling digital printab...

Prompt

I run an Etsy shop called Hajidraft, selling digital printable PDFs. My AI assistant (called Muse) does my niche research. I want you to design a BETTER way to find profitable niches than what we are doing now, and to answer the hard problems listed at the end. Read the whole brief first — do not suggest anything we have already tried without explaining what is different about your version. ## 1. The goal Find Etsy keywords/niches that meet ALL of these at once: - HIGH absolute demand: roughly 20,000+ searches per 12 months, ideally more. - LOW absolute competition: fewer results than the demand justifies. Our working bar is fewer than ~8,700 total results at 20k+ searches, or a demand-to-results ratio above 3.0 at 15,000+ searches. - Conversion badge High or Very high (Etsy's own badge — see section 3). - The top-selling listings for the keyword must be mostly DIGITAL products (downloadable files), not physical goods. - Median purchase price at least ~$1.80, so a $1.99–$2.99 single and a higher-priced bundle can make money. - A coherent printable product we can actually build (planners, trackers, checklists, templates, worksheets — PDFs, not physical goods, not software, not services). Ratio alone is NOT enough. A term with 8,700 searches and 2,900 results is too small for me, and a term with 39,700 searches and 27,300 results is too crowded. I want both sides strong. ## 2. My assistant's environment and capabilities (what it can and cannot do) - Muse is an AI agent with its own Linux computer, a filesystem, Python, and a real Chromium browser. - The browser holds a signed-in Etsy seller session for my shop (Hajidraft). Muse can navigate Etsy, open the seller dashboard, use Etsy search with filters, and open listing pages through browser tasks. It cannot use an Etsy API — there is no API connector and no official API access. Everything comes through the browser UI. - Muse never enters passwords. I take over the browser personally for any login. - Muse can write code, keep research logs in files, and run long research sessions in checkpoints (it typically measures a few terms, reports back, and continues when I tell it to). - Muse can build the actual products once a niche is chosen: vector PDFs (300 DPI, Letter/A4/half-letter sizes, fillable form fields), listing copy, titles, and tags. - Muse CANNOT: bulk-scrape Etsy (that risks bot blocks and rate limits; it must stay near ordinary human browsing volume), access eRank or any third-party keyword tool, download data in bulk from Etsy, or see exact search volume for a term except through the tool in section 3. - Current shop state: one live product line, "Sprinkle & Co." baby shower games — a 50-game bundle plus 14 funnel singles, all published. Singles funnel buyers up to the bundle; the bundle never points outward. Nothing buyer-facing ships without my explicit approval of the exact changes. ## 3. The data tool: Etsy Marketplace Insights (the only real demand data we have) Marketplace Insights is Etsy's own seller analytics tool inside the seller dashboard. For one keyword at a time it shows: - Searches over the last 12 months and the last 30 days. - Total listing results (one exact figure, but note: the figure changes when the date-range view is switched — we use the 12-month header-card figure for verdicts). - A conversion badge: Very low / Low / Typical / High / Very high. This is Etsy's own signal for how often searches turn into sales. - Median purchase price range. - A monthly seasonality curve (which months peak and trough). - A "popular listings" shelf — we audit how many of the top listings are digital vs physical, plus their prices and review counts. - A similar-searches list (related terms with their own search figures). Limits: one keyword at a time, no bulk export, no CSV, no API. Every term costs a page load and a careful full-page read. This is the bottleneck. ## 4. What we have already tried (do not repeat these without a new angle) **Six rounds of Insights sweeps** (finance, wedding, career, teacher, legal forms, home organization, and more). The surviving, verified finalists are frozen as our benchmark table — see section 5. **A 123-term autocomplete harvest + result-count sweep.** We typed seed words into Etsy search, harvested autocomplete suggestions, and counted Digital Downloads results. Six terms came back under 1,500 results. BUT this method was flawed and I rejected its finalists: it measured competition only and never verified demand volume, because autocomplete presence is NOT search volume. Do not propose this method again unless you solve the demand-verification gap inside it. **A 33-term "beat the benchmarks" hunt across 10 fresh clusters** (tax/bookkeeping, ADHD, genealogy, hosting/property, funeral-adjacent, trades, music teaching, language worksheets, pregnancy, trucking). Zero winners. See the two near-misses in section 5. **A 30-term batch** across pet care, moving/new home, college, fertility/IVF, menopause, health logs, RV/camping, chickens, creators, divorce — no winner. **Problem-first hypotheses** (buyer problem → product idea → keyword): surgery procedure-specific, IEP parent binder, elderly-parent caregiver organizer, funeral/estate organizer, esthetician forms, small-landlord packet, nurse report sheets by specialty, postpartum recovery. Only the nurse cluster survived, and only provisionally. **Traps we have already fallen into — do not walk us into them again:** - Trademarked demand: a misspelled Disney term showed a low result count, but the demand was for Disney IP. "Rema" turned out to be reMarkable (a trademarked tablet brand). "Coozie" traces to the KOOZIE trademark. - Physical-product demand: "wedding planner book" had a 5.86 demand-to-results ratio but buyers want physical books at $33–$41. "Pregnancy journal" had a 6.29 ratio and a High badge, but only ~3 of 20 top listings are digital — the rest are physical journals at $35–$43. - Small-market winners: a 3.00 ratio on 8,700 searches/yr is not enough absolute demand for me. ## 5. The benchmark table (frozen — these are the numbers to beat) | Keyword | Searches / 12 mo | Results | Ratio | Badge | Median price | |---|---|---|---|---|---| | nurse report sheet | 25,600 | 8,700 | 2.94 | High | $1.80–$2.20 | | end of life planner | 24,900 | 3,800 | 6.55 | Typical | $3.60–$4.40 | | debt payoff tracker | 33,500 | 21,000 | 1.60 | High | $4.50–$5.50 | | obituary template | 39,700 | 27,300 | 1.45 | High | $1.80–$2.20 | | job application tracker | 8,700 | 2,900 | 3.00 | High | $1.80–$2.20 | Notes: nurse report sheet is PROVISIONAL (demand verified twice, but page 1 is ~90% direct competitors). End of life planner is UNVERIFIED (best raw numbers, but Typical badge and never rechecked/audited). "Better than these" means beating this table, not just beating one row. The two closest near-misses from the latest hunt: "airbnb welcome book" — 38,500 searches, 8,400 results, ratio 4.58, $4.50–$5.50, ~16/19 listings digital — but the badge is LOW. "Pregnancy journal" — 87,400 searches, 13,900 results, ratio 6.29, High badge — but physical-dominated (see section 4). Also notable: "adhd planner" — 156,000 searches, but a Low badge and 64,600 competing listings. ## 6. My standing rules for any research you propose - Every verdict must cite recorded numbers. No guessing, no invented search volumes. - Never declare a winner just because it is the best in a batch. - Demand evidence and competition evidence must be kept separate — a low result count is not proof of demand. - Separate total results from directly competing digital products; audit the top listings (price, format, reviews, differentiation). - Ratio is never a standalone pass. - Unverifiable = marked unverified. Insufficient evidence = provisional candidate, not a confirmed winner. - No product or listing is built until I review and approve the research. - No trademarked terms, no invented products without adjacent-demand evidence (label measured vs hypothesis, never "proven"). ## 7. The difficult problems I want you to solve Answer each one concretely, for THIS tooling — not generic Etsy advice: 1. **Speed.** Insights is one keyword at a time. How do we find candidates WITHOUT burning hours per keyword? What pre-filters, applied BEFORE Insights, actually predict a good keyword — and how would my assistant run them with only a browser, files, and Python? 2. **Is the target real?** Does high demand + low competition + High badge + digital fit even co-occur on Etsy with any regularity, or are we hunting something that structurally barely exists because sellers flood gaps fast? Use the evidence in sections 4–5 to argue your answer, and tell me honestly if the right move is to stop hunting and build the best of the five. 3. **The badge problem.** Why do so many high-demand terms carry Low or Typical badges, and can a digital printable win in a Low-badge market anyway? Take "airbnb welcome book" (38.5k searches, 4.58 ratio, Low badge) and argue both sides. 4. **The physical trap.** How do we pre-screen whether a term's demand belongs to physical products BEFORE spending Insights time on it? What signals (price ranges, listing titles, category mix) give it away early? 5. **Demand signals without eRank.** eRank is unavailable. What other legitimate demand signals can my assistant actually access through its browser — Google Trends, Etsy autocomplete volume patterns, competitor review velocity, Pinterest, Reddit, Amazon — and which of them are actually predictive for Etsy digital sales vs noise? 6. **Throughput.** Can the assistant safely run multiple Insights checks in parallel, or does Etsy's session and anti-bot behavior make that risky? What is a realistic, safe number of verified terms per hour or per day with this setup? 7. **Discovery angles beyond keywords.** What non-keyword-first methods could work: mining competitor shop reviews for unmet needs, tracking fast-rising terms (e.g., a term growing ~20% in 30 days), seasonal windows (the next holiday 90 days out), adjacent niches to my existing baby-shower line, or buyer problems that have no obvious keyword yet? Rank these by expected value for my setup. 8. **New clusters.** Propose specific new keyword clusters or angles NOT in sections 4–5, with your reasoning for why each one might clear the bar (demand story + competition story + why it should be digital + why the badge might be High). Do not re-list anything already measured. 9. **Trademark pre-screening.** How do we cheaply filter out trademark traps before deep research, without a full USPTO search per term? 10. **A concrete protocol.** Give me a day-by-day hunting protocol for my assistant: pre-filter steps, Insights verification steps, pass/fail thresholds, when to stop and build, and how to decide between "keep hunting" and "build the best of the five." Make it executable by an agent with a browser, files, and Python — no API, no eRank, no scraping at scale. ## 8. What I want back from you 1. A better discovery strategy than "type random keywords into Insights," tailored to the tooling in section 2. 2. Your answers to the ten hard problems in section 7, with reasoning — and say clearly where you are guessing vs where you have evidence. 3. Your new cluster list (problem 8) with reasoning per cluster. 4. The executable protocol (problem 10). 5. Do not invent any numbers. If you mention a keyword's demand, say whether it is measured, estimated, or a hypothesis — and if it is a hypothesis, say what would verify it.