Source: raw/Claude_Just_Changed_SEO_AI_Search_Forever.md (main walkthrough); raw/How_to_Connect_Claude_to_Google_Search_Console_and_Analytics.md (Composio setup detail)
Creator: Eliot Prince URL: https://www.youtube.com/watch?v=setxOzrPVkU (main) · https://www.youtube.com/watch?v=9J_m6uUSFBI (GSC/GA setup) Duration: 43:32 · 6:51 Platform: YouTube Published: 2026-10-22 · 2026-08-15
Eliot Prince, who says he did SEO for about 10 years, walks through how he runs SEO and AI search for a one-month-old site (halfdaythursdays.com) entirely inside one Claude project. He wires in three data connectors, writes the project’s instructions, has Claude build an “about my business” context document from the site, then runs a buyer-question gap audit, a full SEO and AI-search audit, a 30-day strategy, a week-one execution pass and a live dashboard. Every output goes back into the project as a context file, so later prompts build on earlier ones. The October video was recorded after the September 2026 chat/Cowork merge, and it uses Claude Docs and live artifacts. The prompts themselves are in his AI Recipe Vault; this article records the structure and the decisions behind it.
Key Takeaways
- Connect data first. Eliot calls this the step “most people skip”. Ask Claude for an SEO audit with no data sources and “it probably just ends up making things up”. His stack has three connectors. Composio covers Google Search Console and Google Analytics, which had no native Claude connector when he recorded either video. DataForSEO supplies keyword volume, competition, CPC, backlinks, LLM mentions and page speed. Firecrawl (a native connector) handles deeper page and document extraction than standard web search.
- Name the connector in the prompt. Both videos phrase requests as “go into my Google Analytics through Composio and tell me…”. He says you have to tell Claude which app to reach through Composio, or save “if I say Search Console, always go through Composio” to memory. If the account has several properties, name the site.
- One project holds the whole job. The project’s instructions define the role (an SEO and AI-search advisor for this domain) and its rules: use real numbers, read the live page first, rank fixes by payoff, explain jargon. Each step’s output is saved back with Add to project. Eliot notes this used about 1% of project capacity. Because the context lives in project files, he starts the strategy step in a new chat and loses nothing.
- The context document comes before any SEO work. A positioning prompt written in the persona of April Dunford (author of Obviously Awesome) reads the site page by page, drafts the document and lists what the site doesn’t say. Eliot fills the gaps by dictating answers covering price, schedule, customer, alternatives, proof, start date, policies and markets.
- Buyer-question gap audit. A prompt reads the site “as a first-time buyer” and lists the questions a buyer would have under five headings: cost, problems, risk, comparisons and alternatives. It also asks what an AI assistant would conclude from the site, and returns one table ranked by how likely each gap is to lose business. Eliot’s top gap was price. Competitors publish theirs, while his sits behind the payment page.
- Full audit, about 10-20 minutes. This prompt combines Search Console, Analytics, DataForSEO and web data. It returns a red/amber scorecard (quick wins, on-page, technical, AI search), a top-10 fix list with instructions, and a record of which brands AI assistants named instead of his. The headline finding was that “every search that finds you already knows your name” — the traffic was almost all branded search.
- The strategy separates money pages from blog clusters. Money pages target purchase-intent terms, such as “AI community for small business owners”. Blog clusters build topical authority. Each item is scored on business value, difficulty and demand, and the output becomes a weekly sequence plus a 30-day plan, delivered as an editable Claude Doc. Eliot reports a competitor with 93 referring domains against his 2.
- Have Claude write the finished assets. The week-one prompt asks Claude to write the actual title, hero copy, FAQ, internal links and structured data. Claude worked out that the site was built in Replit and wrote a ready-to-paste prompt for that tool. Eliot switches his Replit connector off during this step because “it has a habit of auto publishing stuff”, and switches it back on once he has reviewed the output.
- Feed in original material before Claude writes the blog posts. Eliot’s own blog-writer skill interviews him first, or takes a transcript (for example his 2.5-hour Claude skills course), so the posts carry insight other sites don’t have. That matches what the wiki’s information-gain research rewards.
- Finish with a live dashboard, not a static report. Ask for a “live artifact in this project”: one screen, every number labelled with its date range, top-10 queries, biggest movers, top pages and progress against the 30-day plan. A side panel shows where each number came from, and the artifact can be pinned and refreshed. The build took 7 minutes.
The seven steps
| Step | What Claude does | Output saved to project |
|---|---|---|
| 1. Connectors | Composio (GSC + GA), DataForSEO, Firecrawl authorised | n/a |
| 2. Project + context doc | Instructions set the role; Dunford-persona prompt builds “about my business” from the site, asks gap questions | About-my-business doc |
| 3. Buyer-question gaps | First-time-buyer read across cost, problems, risk, comparisons, alternatives, plus the “AI read”; pulls DataForSEO competitor data unprompted | Question-gap audit doc |
| 4. SEO + AI search audit | Search Console, Analytics, technical, GEO checks, “who’s named instead of me” prompts | Audit doc + top-10 fixes |
| 5. Strategy | Money-page plan, topical map (three clusters in his case), priority scores, weekly sequence, 30-day plan | Strategy (pasted in as text, see Open Questions) |
| 6. Execute week one | Writes finished copy, FAQ, links and schema; adapts to the site builder (Replit prompt; WordPress “might have worked in a different way”) | n/a |
| 7. Dashboard | Live artifact pulling connector data with source attribution | Pinned live artifact |
How this squares with the wiki’s AI-citation research
Eliot’s video is one operator’s process on one month-old site, not a measurement study. Checked against the AI SEO hub cluster:
- “Fix SEO first” agrees. His order (fundamentals, then technical, then GEO) matches Zyppy’s “Win SEO first” finding that organic rank is the #2 predictor of AI citation (Zyppy ranking factors).
- Referring domains agrees. His “more websites linking to us” fix lines up with SE Ranking weighting referring domains highest for ChatGPT, as summarised in AI Answer Engines Compared. That article also notes authority signals disagree by engine.
- Structured data: no contradiction, but don’t expect much from it. Eliot describes schema as a “little coded business card” that AI and search engines read, and Claude offered to write it as fix #9. Ahrefs’ causal study (Ahrefs schema study) found no AI-citation lift from adding JSON-LD to pages that were already cited. Whether schema helps a page that isn’t cited yet, which is Eliot’s situation on a new site, has no causal test in the wiki. So schema is reasonable baseline work, but it shouldn’t be counted on to win AI citations.
- The question-gap audit partly overlaps FLUQs. Eliot’s buyer questions are explicit ones (price, comparisons, risk). FLUQs targets the latent, unasked ones. Running both covers more ground.
- Search Console data can’t show AI Overview citations. The wiki records that GSC does not report whether you are cited inside an AI Overview (AI Answer Engines Compared, monitoring section). Eliot’s AI-search read relies on the prompts Claude ran plus DataForSEO’s LLM-mention tools. For anything you need to track, use a fixed prompt panel and run each prompt several times.
- Unsourced stat. Eliot cites a figure “something like 95%” of owners believing their site is complete against 80% of consumers saying sites lack information. He gives no source, and this wiki has not verified it.
Try It
- Set up Composio. At
dashboard.composio.dev, stay in the For You view (“Platform” is the developer side), connect Google Search Console and Google Analytics, then add Composio to Claude as a custom connector (see Connectors). Test it with: “Through Composio, get the last 30 days of Search Console overview stats for.” - Add DataForSEO as a custom connector. The MCP URL is on dataforseo.com; Eliot says new accounts get about $5 of free credit. Test with a single keyword-volume question before running the audit.
- Create one project per site. Paste in role instructions, build the about-my-business doc from the live site, and answer the gap questions honestly.
- Run the question-gap audit, then the full audit. Save each result to the project before starting the next step.
- Ask for a strategy limited to your real capacity (he set three posts a week) and a 30-day plan. Then run week one with any publishing connector switched off until you have reviewed the output.
- Build the live dashboard last, and ask for every number to show its date range and data source.
Implementation
Tool/Service: Claude projects + Composio (GSC/GA middleware) + DataForSEO MCP + Firecrawl connector; optional site-builder connector (Replit in the demo). Setup: Customize → Connectors → Add → Add custom connector → paste the vendor’s remote MCP URL → name it → Add → Connect (OAuth). Firecrawl is in the native directory. Cost: Composio is described as free to use. Eliot reports DataForSEO gives about 50 top-up “is going to last me nearly 3 years” at his usage (his estimate). He calls Firecrawl “free to start”. None of these prices were verified for this article. Integration notes: Claude Docs outputs could not be added to a project with the top-right button in his session. He copied the text and pasted it in as text content instead. The audit step took 10-15 minutes and the dashboard about 7 minutes in his runs.
Open Questions
- Does Claude now have a native Google Search Console or Google Analytics connector? Both videos (August and October 2026) say no. Re-check the connector directory before setting up Composio.
- Docs-to-project friction: is the missing “add to project” on Claude Docs a limitation of the September beta, or by design?
- Results: the video shows the plan, not the ranking outcomes. A follow-up with before/after Search Console data would show whether the workflow works.
- Skipped: Eliot’s paid community “HDT SEO plugin” (15 skills) is pitched mid-video and not documented here.
Related
- Claude Connectors — custom-connector setup and the Composio/DataForSEO entries
- Claude as an Autonomous WordPress SEO Manager — the scheduled, write-access version of this workflow
- Claude Projects — the context container the workflow depends on
- Cowork Live Artifacts — Eliot’s earlier dashboard recipes, including SEO Pulse
- Zyppy AI Citation Playbook — a research-backed audit sequence to compare against
- Ahrefs Schema Study — what schema does and doesn’t do for AI citations
- SEO Audit Skill — a skill-packaged alternative to the prompt-by-prompt audit
- AI Recipe Vault — where the prompts live