AI agent workflow: Create a DataForSEO SEO opportunity agent

Define the business problem before collecting keywords

A DataForSEO opportunity agent helps a team decide where its next SEO effort belongs. Give it a product area, the customer problem that area solves, the intended market, and a list of existing pages. A competitor’s high-volume keyword is not automatically relevant to your product, and publishing a second page for the same intent may create avoidable overlap.

Choose a small competitor set and a stable country and language. Set exclusions for unrelated products, jobs, support requests, and branded terms when appropriate. Ask for a ranked shortlist rather than a spreadsheet of every keyword the API can return. The decision should be publish, refresh, investigate, or leave alone, with a reason for each candidate.

Combine database research with a current SERP check

DataForSEO’s Domain Intersection endpoint can return keywords a competitor ranks for while the target does not. Its documentation describes a periodically updated dataset, so record the available update context rather than calling every result real-time. Keep organic results separate from paid placements when the objective is organic search.

For a shortlist, inspect live search results using a fixed location, language, and device. Look at the kinds of pages ranking: product pages, comparisons, tutorials, directories, or local results. Treat that as evidence about the searcher’s likely task, then compare it with what your page can honestly deliver. Where available and approved, AI Overview observations can add context about displayed answers and citations. One capture does not establish stable inclusion, exclusion, or share of voice.

Example starter prompt

Use DataForSEO to find SEO opportunities for [domain and product area] in
[country, language, and device]. Compare [competitors] and inspect these
existing pages: [URLs]. Exclude [unrelated topics and branded terms].

First inspect available tools and propose a bounded request plan with current
cost information. Ask before paid calls, deeper SERP collection, or optional
AI Overview retrieval. Stay within [budget and keyword limit].

For the best candidates, return query, intent evidence, ranking page types,
existing target page, relevant provider metrics, collection context, and a
publish/refresh/investigate decision. Label estimates and unavailable fields.
Draft a brief for the top three decisions. Do not publish or change the site.

Make the recommendation useful to an editor

A refresh brief should identify what the existing URL fails to explain and what evidence would make it more useful. Suggest concrete changes such as a worked example, corrected product detail, clearer comparison, or better alignment between title and page purpose. Do not copy competing pages or add unsupported claims merely because they appear in a search result.

For a proposed new page, explain why an existing URL cannot serve the intent. Include an outline, required subject-matter input, internal-link candidates, and a measurement plan. Estimated traffic, search volume, and advertising competition are provider metrics, not observed visits, organic difficulty scores, or guaranteed conversions. If first-party performance data is supplied separately, preserve its definitions and permissions.

After approved implementation, retain the URL, change date, and baseline query settings. Revisit the same scope and distinguish search-result changes from actual business outcomes. Technical blockers found during the research can move into the Linear SEO workflow.

Apply an approved refresh in the project repository

The research does not have to end in a document. In a LatchLoop project connected to the website repository, ask the project agent to inspect the relevant article files, content schema, layout, and existing internal links. LatchLoop can edit those files directly; DataForSEO does not need a website-writing endpoint for that step. The background build agent guide explains the implementation workflow.

Keep the change scoped to the approved brief. The agent can update an article’s examples, headings, title, description, and relevant links, then run the repository’s checks and prepare a diff for review. Preserve URLs unless a redirect plan is approved. Canonicals, structured data, and indexing directives need separate scrutiny because a broad template edit can affect more pages than the research covered.

Use the approved SEO brief above to refresh [article URLs] in this project's
repository. First locate the source files and follow the site's content schema.
Edit only the approved copy, metadata, and internal links. Preserve factual
qualifications and cite sources. Do not invent product claims or add keyword
repetition just to match the brief.

Run the relevant content checks and build. Return changed files, a summary of
the reader benefit, test results, and preview-review steps. Do not merge or
publish without review. Keep the baseline and measurement plan with the task.

For content stored outside the repo, a write-capable plugin such as WordPress.com can provide the destination instead. Inspect its actual tools, identify the exact post, and review old and proposed values before approving a draft update. A content save and a public publication are different actions. Neither a repository change nor a CMS update proves SEO improvement; measure the deployed result against the baseline.

Questions this workflow answers

Which SEO opportunities deserve attention when keyword tools return thousands of suggestions?

Filter for the business and the searcher’s task before ranking by volume. A smaller query with a clear connection to the product may deserve attention before a broad topic that attracts the wrong audience. Inspect the result types to understand whether people want to buy, compare, learn, troubleshoot, or find a local service. Then check whether the site already has a page that can satisfy that need with a focused improvement. This reduces the temptation to publish a new article for every variation of the same question.

Use a short decision table with explicit tradeoffs: relevance, observed intent, existing coverage, effort, evidence gaps, and measurement plan. A large provider estimate should not conceal weak fit or missing expertise. For current AI-search features, preserve the query and observation conditions and avoid promising a citation outcome. Ask the content owner to approve the top few briefs, implement those changes, and evaluate them against a recorded baseline. The agent’s contribution is a defensible work queue and better questions for the team, not an automatic ranking strategy that bypasses editorial judgment.