The useful SEO prompt is not “optimize everything”; it is a small implementation brief tied to evidence from the live website.
Give the coding agent the production URL, exact failing evidence, expected behavior, owning route or component if known, allowed files, protected routes, content facts it may use, claims it must not invent, and executable acceptance tests. Ask for the smallest compatible change, syntax and project tests, a concise change summary, and post-deployment verification steps. Do not ask it to generate dozens of keyword pages, expose private routes, or claim rankings and citations.
This guide provides a reusable prompt structure for technical, on-page, AEO, and GEO implementation while preserving architecture, user data, route boundaries, and honest outcome language.
Main Explanation
Start with observation. Include the requested production URL, final status, redirect chain, relevant HTML or header, screenshot or audit finding, and why it blocks a user or crawler. Distinguish what was measured from what is inferred. “The canonical points to the preview domain” is actionable; “Google hates the site” is not.
Define boundaries. Name the intended public routes, private routes that must remain protected, files or components the agent may change, framework or plain architecture to preserve, and unrelated behavior it must not rewrite. If the page copy needs factual changes, supply approved product facts and primary sources. Prohibit invented statistics, reviews, integrations, guarantees, or customer claims.
Specify acceptance tests before implementation. Examples include one H1, unique metadata, a self-referencing canonical on the preferred host, a valid sitemap containing the route, a real 404 for missing paths, useful text in source output, accurate visible schema, and successful anonymous access. Require local syntax or build tests and a production retest after deployment.
Use one prompt per dependency. Fix access and status before content polish; fix canonical and sitemap consistency before submitting URLs; establish a useful canonical page before adding schema. Small prompts produce changes that are easier to review, roll back, and attribute than a single request that mixes architecture, content, design, and measurement.
Keep the public acquisition layer separate from the private application. Marketing, use-case, comparison, documentation, support, and policy pages may be discovery targets when they serve visitors. Account, dashboard, billing, checkout, API, webhook, admin, preview, staging, test, and user-data routes should remain protected and outside public sitemaps. A crawler directive is not a substitute for authentication or authorization.
Turn each finding into a bounded implementation task. Record the production URL, observed response or element, affected route scope, owning layer, files or settings allowed to change, routes that must remain private, and the acceptance test. After the smallest safe correction, run project checks, publish, and repeat the exact anonymous request. A builder preview or local success does not prove the production response changed.
Measure the stages separately. A successful fetch is crawl evidence, Search Console reports Google discovery and performance, infrastructure logs record requests, analytics records attributed visits, and controlled observations or provider reports can record AI appearances. Technical readiness can remove blockers, but no builder, prompt, sitemap, schema block, scan, or indexing request guarantees ranking, citation, recommendation, traffic, or conversion.
Practical Steps
- Capture the production evidence.
- Name the desired behavior.
- Define allowed scope.
- Protect private and unrelated routes.
- Provide approved facts and sources.
- Write executable acceptance tests.
- Request the smallest compatible change.
- Deploy and repeat the original test.
FAQ
Can I paste a full SEO audit into one prompt?
Use the audit as context, but implement one dependency or tightly related group at a time so the change remains reviewable.
Should the agent create content automatically?
Only from approved facts and a distinct visitor intent, with human review for accuracy, usefulness, and overlap.
What is the most important line in the prompt?
The acceptance test: it defines how you will prove the production issue was actually corrected.
Sources and methodology
Reviewed on 2026-08-26 against current first-party platform documentation and Google Search Central guidance. The workflow evaluates the public production response and keeps crawlability, indexing, search performance, AI visibility, referrals, and business outcomes separate.
These references support the changeable facts and study findings discussed above. Results depend on each source's sample, date, market, query set, and measurement method.