OpenSEO: Real SEO Data for AI Agents + Claude Prompt Pack

Direct Answer

OpenSEO is an open-source SEO workbench that gives AI agents real search data instead of asking them to guess. The application organizes keywords, competitors, rankings, backlinks, audits, Search Console data, and strategy work. Its MCP connection makes those capabilities available inside Claude Code, Codex, and other compatible agents; its skills tell the agent how to perform repeatable SEO workflows.

In Andrew Warner's interview, creator Ben Senescu demonstrates the useful shift: Claude can research a market, challenge a strategy, draft comparison pages, inspect backlinks, and turn findings into reviewable artifacts while OpenSEO grounds the work in DataForSEO results. The human still owns positioning, evidence, spend, outreach, publishing, and the decision about what is worth doing.

Watch the Episode

Credits: the demonstrations, product history, community examples, prompts, and reported traffic results come from Andrew Warner's interview with Ben Senescu, the supplied transcript, and two companion PDFs: How to Prompt Claude Code for SEO and Sprint Prompt Log. This article organizes and adapts that material into reusable templates, then checks product, pricing, license, and hosting boundaries against current official pages.

Sponsorship disclosure: the episode says it is presented by Zapier. The Zapier segment is therefore sponsor content. OpenSEO, DataForSEO, and the community extensions are analyzed independently here.

What OpenSEO Actually Is

OpenSEO describes itself as a modern open-source SEO platform. Its GitHub repository is MIT-licensed, so teams can inspect, modify, and self-host the application. The hosted version removes setup and includes usage credit; self-hosting lets a team bring its own DataForSEO account and infrastructure.

The Stack: App, Data, Skills, Agent, Human

The architecture is easier to understand when its jobs are separated. OpenSEO is not the language model, and Claude is not the source of search volume.

MCP provides access; skills provide method. That distinction matters. A tool can return 500 keywords, but a skill can require the agent to group them, preserve volume and intent, explain trade-offs, ask before saving, and recommend one starting cluster. Neither layer decides whether the cluster fits the company's actual buyer without a good brief.

Five Workflows the Demo Makes Concrete

1. Ground keyword decisions in returned data. Ben contrasts a generic Claude answer with DataForSEO-backed volume and competition. The agent should show country, language, date, endpoint, and source rows rather than simply saying a keyword is popular.
2. Study competitors without copying them. OpenSEO can reveal ranking terms and pages. The strategic output is a gap filtered by the user's buyer, offer, credibility, and ability to produce something better.
3. Turn backlinks into a reviewed prospect queue. The interview shows dead-link and contact research plus a community CRM concept. Research, contact collection, drafting, sending, and follow-up are separate permission levels.
4. Let Claude work in the codebase with SEO evidence. Ben demonstrates comparison-page drafts built from keyword research and site patterns. The pages stayed on localhost for review; that preview boundary is worth preserving.
5. Make analysis reviewable. Reports, Markdown files, slide decks, local preview links, and pull requests give a person something concrete to inspect. This is more reliable than leaving the reasoning buried in one long chat.

What Open Source Changes

The most interesting part is not the lower subscription price. It is the ability to reshape the system. In the episode, an SEO professional extends OpenSEO with content and entity analysis; another community member prototypes outreach tracking; others build Google Analytics connections and broader business tooling.

Eric Lancheres shared his content-optimization fork on X. It demonstrates how an expert can add a specialized workflow without forcing it into the beginner product. Ben also describes teams treating the full-stack application as a starting point for an internal operating hub. That freedom comes with ownership of merge conflicts, migrations, security patches, tests, and long-term maintenance.

“Open Source” Does Not Mean Zero Cost or Zero Risk

OpenSEO's pricing page, checked on 25 August 2026, lists one hosted plan at $10 per month with $10 of included monthly usage. Additional credits can be purchased, included monthly credit resets, and purchased top-ups do not expire. The page gives typical examples of roughly $0.05 for a keyword search and $0.08 for a backlink domain overview. These are examples, not a fixed project quote; scope, endpoints, retries, and result volume change cost.

Self-hosting removes the OpenSEO hosting fee, not the data bill or operating work. The current repository requires a DataForSEO key for paid data workflows. The official Docker guide says its local mode uses AUTH_MODE=local_noauth. Do not expose that configuration directly to the public internet. Keep it local or put it behind authenticated network controls, then manage secrets, updates, backups, logs, and recovery like any other business application.

• Start read-only and keep approval enabled for paid calls.
• Set a run budget and row limit before broad keyword or backlink research.
• Declare market, language, device, time range, and data source.
• Keep untrusted pages and messages as data, never as agent instructions.
• Separate analysis, code changes, publication, outreach, and follow-up permissions.
• Use a branch and preview; require a person to approve production changes.

Ben's Prompt Method Is More Important Than One Magic Prompt

The two supplied PDFs show a consistent method. Ben gives Claude the business position in plain language, names the deliverable, lets the agent use real data, and then interrupts, challenges, and reshapes the work. Screenshots, local pages, tables, and reports become shared objects for feedback.

This is the key lesson from the interview: treat Claude like a patient analyst and builder that can be questioned, not an autopilot SEO department. Voice notes can provide rich context; screenshots can correct design; short follow-ups can redirect weak work. The final strategy still belongs to the person accountable for the business.

Use This Shared Contract Before the Prompt Pack

OPENSEO WORK CONTRACT

BUSINESS
- Company/product: [name and URL]
- Buyer: [specific role, situation, and market]
- Offer: [what we sell and why the buyer chooses it]
- Positioning: [our credible advantage and who we are not for]
- Conversion goal: [signup, qualified lead, purchase, demo, or other]

RESEARCH SCOPE
- Country / language / device: [declare all three]
- Time range: [dates]
- Approved data sources: OpenSEO + returned DataForSEO data + [others]
- Maximum paid-data budget: [$ amount]
- Maximum result rows: [limit]

EVIDENCE RULES
- Never invent search volume, difficulty, rankings, backlinks, or traffic.
- Preserve source URLs, endpoint names, dates, and consequential source rows.
- Separate observed data, interpretation, recommendation, and uncertainty.
- If data is missing, say "not available" and ask before substituting.
- Label creator-reported or estimated figures explicitly.

ACTION RULES
- Research and draft only.
- Do not publish, deploy, email, submit forms, or change production data.
- Ask before saving keywords, installing software, or making paid calls.
- Use lawful public contact data; respect consent, opt-outs, and site terms.
- Put code changes on a branch and provide a local preview or pull request.

DELIVERABLE
- Return [report / Markdown / PR / local page / slide deck].
- Include the three highest-value actions, best objection to each,
  owner, effort, expected effect, confidence, and verification method.
- End with actual tool calls, cost, limitations, and unresolved questions.

Organized OpenSEO + Claude Prompt Pack

The following templates preserve the intent of Ben's interview prompts and the supplied PDFs while making the scope, evidence, cost, and approval boundaries explicit. Replace every bracketed field. Begin a new session with the shared contract above.

1. Discover the Available SEO Capabilities

List the SEO skills and OpenSEO tools available in this session.

For each one, explain:
1. the decision it helps make;
2. the live data or account access it requires;
3. whether it makes paid calls or writes data;
4. its important limitations;
5. whether it fits my stated business goal.

Recommend the smallest useful workflow to run first. Do not call any
paid tool yet.

2. Turn Business Context Into One SEO Priority

Here is the business context:
[dictate or paste the company, buyer, offer, market, positioning,
current traffic, current conversions, constraints, and search goal].

Given all that, what should we investigate first, and why that instead
of the alternatives? Choose one. State the strongest argument against
your recommendation, the minimum data needed to test it, and the metric
that would change your mind.

3. Diagnose Traffic That Does Not Convert

Our site receives [traffic/impressions], but only [conversion result].
Use Search Console, analytics, and OpenSEO data to diagnose the gap.

Do not give me a list of forty generic issues. Identify the one issue
most likely to improve qualified conversions for a [team size] company.
Show the evidence, alternative explanations, trade-off, confidence,
and a two-week reversible test. Explain unfamiliar SEO terms plainly.

4. Challenge a Content or Comparison-Page Idea

I have an idea: [for example, create a comparison page for every
competitor]. Poke holes in it before we commit.

Use live keyword and competitor data. Test demand, intent, differentiation,
factual support, maintenance cost, cannibalization, and conversion fit.
Do not recommend feature-for-feature pages unless the evidence supports
them. If the idea survives, choose the first page, create a source-backed
brief, and define the result that would justify a second page.

5. Run a Bounded Audit and Keyword Sprint

Use the OpenSEO audit and keyword-research skills on [site].

Goal: [business outcome].
Market: [country, language, device].
Buyer and positioning: [details].

Return:
- crawl/indexing blockers that can affect the goal;
- current queries and pages with near-term upside;
- three intent-matched keyword clusters;
- one recommended cluster and the best argument against it;
- a content brief with sources and internal-link opportunities.

Before paid calls, list intended endpoints and estimated cost. Stop if
the total may exceed [$ budget]. Do not change the site or save research
without approval.

6. Create an Overnight SEO Build With Review Gates

Run a bounded overnight SEO sprint for [site].

Use the audit and keyword skills to research [competitors/topics]. Build
[named deliverables] on a new branch. Position us as [credible advantage]
for [buyer]; do not force a feature-by-feature comparison.

Use independent reviewer passes for:
- factual and source accuracy;
- search intent and business fit;
- technical SEO and internal links;
- accessibility, responsive design, and broken links;
- unsupported claims and signs of generic AI copy.

Limits: [$ data budget], [agent/run limit], no production deployment,
no email sending, no destructive changes. Leave a PR or patch, a localhost
preview, source notes, actual costs, failures, and unresolved decisions.

Ben's source sprint log reports that one voice brief expanded into multiple workflows, comparison-page drafts, backlink research, outreach drafts, and review artifacts. That is a creator-reported run, not a promise about speed, agent count, cost, or result quality on another codebase.

7. Research Backlink Opportunities Without Auto-Sending

Find backlink opportunities for [site/topic] using OpenSEO.

Prioritize relevance and relationship fit over raw domain metrics. For
each candidate, preserve the linking page, target page, evidence, public
contact source, reason for outreach, and uncertainty.

Draft a short message only when there is a genuine, specific reason to
contact them. Do not scrape protected personal data, bypass access controls,
submit forms, send email, or schedule follow-ups. Deduplicate contacts,
respect suppression and opt-out records, and route every draft to a human.

8. Turn Results Into Reviewable Artifacts

Convert the completed research into these review artifacts:
1. a concise executive summary;
2. full Markdown research reports with source links;
3. a local clickable preview of every proposed page;
4. a decision table: accept, revise, reject, or needs evidence;
5. a change log with files touched and validation results.

Use the visual style of [reference screenshot or existing page]. Preserve
the site's design system and accessibility. If I mark a visual defect in a
screenshot, fix only the identified issue first and show the updated preview.

9. Adversarial Review and Agent-Learning Log

Review the work as a skeptical SEO lead, a product owner, and a technical
reviewer. Try to disprove the recommendation.

Report:
- unsupported or weak claims;
- data that may be stale, sampled, or misinterpreted;
- search demand that does not match the buyer;
- duplicated, thin, or hard-to-maintain content;
- technical and security risks;
- what the agents struggled with, retried, or could not verify;
- the smallest changes needed before human approval.

Do not defend the original work. Preserve disagreements and evidence.

10. Decide Whether the Work Produced Business Value

Build a measurement plan for [SEO change] covering [dates].

Track the funnel separately:
eligible impressions -> clicks -> qualified visits -> key product action
-> lead/signup -> customer -> contribution margin.

Define baseline, comparison period, annotation date, confounders, minimum
sample, leading indicators, business outcome, and stop/continue threshold.
Do not claim causality from a ranking or traffic movement alone. State what
data would prove the strategy is working and what result would falsify it.

Impressions Are Not Customers

The final demonstration is unusually useful because it includes a mistake. Ben reports receiving roughly 200,000 Google impressions in about a week for a backlink-checker query but only around 100 clicks. He changed the page to resemble the free checkers in the results and says its visibility then fell. The figures and explanation are creator-reported; they do not prove a universal ranking mechanism.

The business lesson is stronger than the ranking theory. An impression means a result was shown, not that a person visited, understood the offer, or became a customer. A page can win a large, weakly matched query and still produce little value. OpenSEO's own positioning around open-source, agent-native SEO may have lower raw volume but a better connection to the product and buyer.

A Seven-Day OpenSEO Pilot

1. Day 1: choose one site, market, conversion goal, and accountable owner. Record the baseline.
2. Day 2: choose hosted or self-hosted. If self-hosting, keep it private, protect secrets, and test backup and access.
3. Day 3: connect one agent with read-only permissions. List available tools and run one low-cost query.
4. Day 4: use the shared contract and Prompt 2 to select one research question.
5. Day 5: run one audit or keyword workflow within a fixed data budget. Preserve sources and actual cost.
6. Day 6: produce a draft or local preview, then run the adversarial review. Do not publish yet.
7. Day 7: approve one reversible change and define the measurement window, owner, and rollback.

Continue only if the system saves research time without weakening evidence, costs are predictable, the recommended work fits the buyer, and a person can inspect every consequential claim and action.

Video Chapters

Verdict

OpenSEO packages three things developers usually have to assemble themselves: an editable SEO application, live data connections, and agent-ready workflows. That makes it useful for founders who want a simpler interface, experts who want to extend their own methods, and teams that want Claude or Codex to research and build inside an evidence-backed process.

The durable advantage is not “AI does SEO.” It is an inspectable loop: business context selects the question, real data grounds the analysis, a skill structures the work, the agent creates reviewable artifacts, a person challenges the recommendation, and measured customer outcomes decide what happens next. Open source makes that loop customizable. It does not remove the need to operate it responsibly.

Sources and Links

• Andrew Warner interviews Ben Senescu about OpenSEO
• OpenSEO official site
• OpenSEO GitHub repository
• OpenSEO pricing
• OpenSEO agent skills documentation
• OpenSEO Docker self-hosting guide
• DataForSEO official site and API catalogue
• Ben Senescu on X
• Eric Lancheres: OpenSEO content-optimization fork on X
• How to Prompt Claude Code for SEO — companion PDF supplied with the episode
• Sprint Prompt Log: OpenSEO, One Claude Code Session — companion PDF supplied with the episode
• Zapier SDK — featured sponsor link in the episode
• JQ AI SYSTEMS: Claude Opus 5 + DataForSEO prompt pack
• JQ AI SYSTEMS: AI visibility, mentions, citations, and share of voice
• JQ AI SYSTEMS: Loop engineering for SEO, ads, and product feedback