How to use this tool — seed formulas, templates & best practices
Get the first step right and everything downstream — scoring, ad groups, SEO clusters, website content — comes out right on its own. Give 3–6 seed keywords, one from each pattern shown on the card. Click Use template to fill the form (niche, location, seeds, deliverable type), then adjust it for your business.
Every run, 5 rules: 3–6 seeds · no brand names · no single generic words ("cleaning" ❌ → "sofa cleaning dubai" ✅) · one language per run (do a separate Arabic run — the data comes back in Arabic too) · location as free text ("Lahore, Pakistan", or "N/A" for worldwide) — the geo target resolves automatically
The 5 deliverable modes — every pipeline explained
MODE · GOOGLE ADSlaunch or expand a paid search account
What runs: your seeds go to the real Google Ads Keyword Planner (volumes, CPC ranges, competition, 12-month trend) → Python dedupes, classifies intent and scores 0–100 against the dataset itself → Claude designs 1–3 campaigns and up to 10 ad groups (count comes from the data, never padded) with cross-group negative siloing so groups can never cannibalize each other → every keyword gets a suggested starting bid computed from its real top-of-page bid range → 15 headlines + 4 descriptions per ad group (policy-checked) → audience plan matched from the 3,504-segment master list → an hourly negative-guard Ads Script → Editor-ready geo target rows.
Best when: a client is paying for clicks. Fill Website URL so RSA Final URLs point at real landing-page slugs, never the homepage.
Perfect-use tips: pick country + city in the location picker (city → the Locations CSV targets that city's real areas). Account already live? Fill Existing campaign + ad groups — new groups arrive pre-siloed. Small niche → cap Max new ad groups at 1–2.
MODE · SEOcontent plan for a site that should rank
planner pull → python scoring → claude cluster strategy → one JSON link for the website builder
What runs: same real-Planner data collection and scoring, then Claude builds the Mode 4 cluster architecture: a pillar topic + 3–8 clusters (count data-driven), each with its primary keyword, keyword set, funnel stage, 4–8 real customer questions with answer angles (the pattern that wins featured snippets and AI overviews) and entities to mention. One keyword never sits in two clusters — cannibalization guard. The same JSON also carries a demand-backed services list (Mode 3), a services hub (Mode 2) and city/area targets (Mode 5).
City-level pSEO: pick a city in the location picker and the Mode 5 block is built from that city's real Google geo areas (all candidates included as geo_area_candidates) — ready for per-area pages, no invented area names.
You get: the Mode 4 SEO data link on the result page — paste it into the website builder's SEO data URL field and leave Main topic / Problem clusters blank; they fill from real demand.
Perfect-use tips: question seeds ("how to…", "cost of…") are the gold here. Blogs/affiliate: location N/A for worldwide volumes. Multi-section sites: one run per section, each with its own parent slug.
MODE · BOTHthe default for local service clients
one data pull → the full Ads package + the full SEO package
What runs: everything from both modes above on a single Planner pull — same seeds, same scoring, two strategy stages. The ads keywords and the SEO clusters come from the same demand data, so your landing pages and your content plan tell one consistent story.
Best when: a client wants leads now (ads) and cheaper leads later (SEO) — which is almost every local service business.
Perfect-use tips: BOFU commercial keywords go to ads; TOFU problem/question keywords are auto-excluded from ads but become your strongest SEO pages — nothing is wasted, every keyword lands on the side it converts on.
MODE · MODE 3 SITE PLANfull website — services list in, per-page keywords out
paste your services list → claude groups into categories → per-category planner pull → per-page keyword + question assignment
What runs: the seed box becomes a services list (paste the exact same comma-separated list you give the website builder's Mode 3 — 100+ services is fine). Claude groups them into up to 10 categories (names copied character-for-character), each category gets its own Planner pull for clean attribution, then every page is assigned its primary keyword, supporting keywords with real volumes, and 3–6 questions with answer angles — plus a site-wide cannibalization guard.
You get: one JSON link for the builder's seo_inputs_url field — the builder adopts this plan's category tree and real keywords instead of guessing.
Perfect-use tips: the services list must match the builder run exactly. Zero-volume services stay in the plan (flagged) — pages still build. Use for high-value clients; for quick sites the builder's own guesses are fine.
MODE · LOCATIONS ONLYinstant geo CSV — no pipeline run, no wait
campaign name(s) + country/city picker → Editor CSV built in your browser in seconds
What runs: nothing on the server — your browser fetches the real Google geo-target database and builds the same 18-column Editor CSV the full pipeline produces. Country only → country + all states/provinces. Country + city → the city + its province + every sub-area inside it (Dubai = 107 rows).
Best when: you already have campaigns and just need proper location targeting — enter the exact campaign names from your account, comma-separated for several at once.
Perfect-use tips: import via Ads Editor → Account → Import. Campaigns that don't exist yet import Paused. After import set each campaign to "Presence: people in or regularly in" — never "interest" — so budget only buys clicks inside the target.
🚀 Fastest path — API push (no Editor at all)
Fill the API Push — Google Ads customer ID field and the whole campaign goes straight into the account through the Google Ads API — campaign settings, ad groups, keywords with bids, negatives, RSAs, locations with premium/low bid tiers, excluded locations, and audiences. No CSV, no import, no format errors, and it even sets Presence-only location targeting (which a CSV can't).
Validate first (default): Google checks every single operation and reports any problem without creating anything — account stays untouched. Read the API Push Report on the result page.
Validation clean? Run again with Live — the campaign lands Paused (zero spend) in one atomic write: either the whole thing lands or nothing does.
Requirement: the account must be linked to your MCC (Manager Account). Only the guard script stays manual (Ads Scripts can't be pushed via API).
Leave the field blank and you get the CSV files exactly as before — the manual path never goes away.
Full account import order (Ads Editor)
⭐ Shortcut — Master CSV: one file, one import, complete campaign: campaign settings (Search only, Google partners OFF, Manual CPC, Paused) + ad groups + keywords with bids + negatives + RSA ads + locations with premium/low bid tiers + excluded locations + audiences included AND excluded (via the Editor's documented Type column). Import via Account → Import, set the daily budget, review, Post. The ONLY thing left out is the guard script (Tools → Scripts — Ads Scripts can't import from CSV). Steps 1–5 below are the manual per-file path if you prefer it.
Ads Editor CSV — campaigns, ad groups, keywords with suggested bids.
Negatives CSV — the anti-cannibalization layer. Always a separate import — pasting positives and negatives together makes the Editor import everything as positive keywords.
RSA Ads CSV — responsive search ads per ad group.
Audiences CSV (+), then Audiences Negatives CSV (−) — again, two separate imports.
Locations CSV — geo targets per campaign.
Set budgets and bidding manually, review everything, then Post. All campaigns arrive Paused — nothing spends until you enable it.
Golden rule: every file goes through Account → Import (paste text or select file) — never paste into the keyword grids directly; the grids ignore the Criterion Type column.
Business type templates
1 · LOCAL SERVICEAC repair company — Dubai, UAE
core service · service + city · price/cost modifier · problem phrase · urgency
Niche:AC repair, installation and maintenance company for homes and officesDubai, UAE
ac repair dubai, ac not cooling, ac service cost dubai, emergency ac repair, ac maintenance dubai
• Deliverable: Both — the Ads package (campaigns, negatives, guard script) and the SEO package (Mode 4 cluster site) in one run.
• Problem phrases ("ac not cooling") pull the diagnostic long-tail that becomes your strongest SEO pages.
• Small niche? Set Max new ad groups to 1–2 — the tool never pads to look thorough.
2 · MULTI-TRADE SERVICEShome maintenance company — this card = the plumbing run
run each trade SEPARATELY — clone this run for electrical, carpentry, painting…
Niche:Plumbing repair and installation services for villas and apartmentsDubai, UAE
plumber dubai, emergency plumber near me, plumbing repair cost, leaking pipe fix, water heater installation
• Never mix trades in one run — the 400-keyword cap and the percentile floor squeeze out each trade's long-tail and the clusters go shallow.
• Ads: each run = its own campaign (per-trade budget control). Adding a trade to a live account later? Fill the Existing campaign / ad groups fields — 2-way negative siloing is automatic.
• SEO: each run's Mode 4 JSON builds its own site section via parent slug (/plumbing/, /electrical/). Umbrella homepage → Mode 3.
3 · ECOMMERCE STOREonline perfume store — UAE
buy + product · product + price · best X for Y · category / type · occasion / gifting
Niche:Online store selling perfumes and fragrances with UAE-wide deliveryUnited Arab Emirates
buy perfume online uae, perfume price in dubai, best perfume for men, oud perfume, perfume gift set for women
• One run per main category (perfumes ≠ watches) — one category = one clean cluster architecture.
• Never seed your own brand or competitor brands — competitors get auto-excluded and auto-negatived.
• SEO clusters become category pages + buying guides; "best X for Y" seeds create the MOFU comparison content that converts.
4 · BLOG / CONTENT SITEhome improvement blog — worldwide
how-to question · cost query · best X comparison · DIY topic · ideas / inspiration
Niche:Home improvement blog covering repairs, renovation costs and DIY guidesN/A
how to fix a leaking tap, bathroom renovation cost, best water heater for home, diy wall painting tips, small kitchen design ideas
• Deliverable: SEO — a blog doesn't need the ads package; question/informational keywords are the content gold here.
• Question seeds make the tool generate AI-overview / voice questions with answer angles — they become the H2/H3 headings on your pages.
• Monetizing with affiliate? Add "best X" comparison seeds — MOFU clusters are where the revenue is. Location N/A = worldwide data.
Niche:Real estate agency for buying, selling and renting apartments and villasDubai, UAE
apartments for rent dubai, buy villa dubai, studio for rent in marina, property prices dubai, off plan projects dubai
• Rent vs buy are different intents — expect the tool to silo them into separate ad groups with cross-negatives automatically.
• Area names in seeds ("marina") feed Mode 5 pSEO: the JSON lists areas ranked by search demand for per-area pages.
8 · EDUCATION / COACHINGIELTS academy — Karachi, Pakistan
course + city · fees / price · best X · online variant · how-to (content)
Niche:IELTS and English language test preparation academy with online classesKarachi, Pakistan
ielts preparation karachi, ielts coaching fees, best ielts academy, ielts online classes pakistan, how to prepare for ielts at home
• Use the words your market uses — here "fees" is the price modifier, not "price".
• Serving remote students too? Do a second run with location N/A (worldwide) for the online demand — different volumes, different competition.
• "how to prepare" seeds feed the SEO content plan (TOFU articles that rank and funnel into the course pages).
Google Ads — best practices after the run
Import the CSVs: download Ads Editor CSV (positives) from the result page → Google Ads Editor → Account → Import. Campaigns, ad groups, keywords (Phrase/Exact — never Broad) and intent expansions land in one import. Then import Negatives CSV as a separate second import — never mix the two files (see the import-order box above). Review, then post.
Install the guard script: Google Ads → Tools → Scripts → paste the Negative Guard Script → run with DRY_RUN = true for 2–3 days → read the logs → set it to false and schedule hourly. Search terms with conversions are never banned.
Build matching landing pages:landing_pages.json (in the run artifact) gives ready Mode 1 generator inputs per ad-group theme — message match raises Quality Score and lowers CPC.
Account already running? Fill Existing campaign name + Existing ad groups before the run — new groups slot into that campaign with 2-way negative siloing, plus a list of negatives to add to your old groups.
Who decides how many ad groups? The data does (one group per genuinely distinct commercial theme — if it only supports 4, you get 4), you cap it (Max new ad groups), and Python enforces it (every keyword in exactly one group, empties dropped).
SEO — best practices after the run
Copy the link: the result page shows the Mode 4 SEO data link the moment the run is ready — no manual file handling.
Feed the website builder: choose Mode 4 → paste the link into SEO data URL → leave Main topic and Problem clusters blank; they fill themselves from real search demand.
What each page gets: its own primary keyword (forced into the hero title and first paragraph), the top-10 target keywords with volumes, and real customer questions as H2 sections and H3 FAQs — each answer's first line follows its answer angle, the pattern that wins featured snippets and AI overviews. The pillar page gets the pillar keyword and its own questions.
Multi-section sites: one run per section, each with its own parent slug (/plumbing/, /electrical/).
Same JSON, other modes: it also carries a demand-backed services list (Mode 3) and real geo-area city targets (Mode 5 — pick a city in the location picker and every candidate area ships in geo_area_candidates) — reuse the same link when you run those modes.
Cluster count is data-driven: 3–8, based on the distinct problems actually visible in the keywords — never padded, and one keyword never appears in two clusters (cannibalization guard).