AI Visibility Audit: Help ChatGPT and Claude Understand Your Website

Start with the right model

AI readability is necessary, but it is not a recommendation guarantee

An assistant can only use public information it can retrieve or access through its search and browsing systems. Clean responses, crawl permissions, readable HTML, clear content, and normal links reduce technical friction.

Recommendation is a larger problem. Relevance, product fit, authority, freshness, independent mentions, available search indexes, user context, and the assistant’s own retrieval system can all affect whether your company appears in an answer.

↳

A higher readiness score means the scanner observed fewer technical barriers. It does not prove that ChatGPT, Claude, Gemini, Perplexity, or another assistant will cite or recommend the site.

Step 01

Scan the deployed website-not localhost

Open Is-Agentic, enter the canonical public URL, and select Score. The service says its scan is run by Ora and evaluates what agents can discover, access, understand, and use on the public site.

Use the production URL because local development does not reproduce CDN rules, bot protection, redirects, response headers, robots directives, authentication boundaries, or deployed rendering behavior.

• Confirm the report scanned the intended canonical URL after redirects.
• Save the scan time and current score as a comparison point.
• Read the observed evidence for each failed or partial check.
• Do not start with bonus points while Essential checks remain broken.

Step 02

Work through Essential, Recommended, then emerging checks

The tool currently assigns most of its score to Essential checks. These cover baseline behavior such as server-rendered content, correct HTTP responses, document structure, recoverable errors, usable controls, canonical URLs, language declarations, and Open Graph metadata.

• Essential: barriers that can prevent an ordinary agent from reliably accessing or interpreting the site.
• Recommended: checks activated when the scan detects relevant capabilities such as an API, OAuth flow, GraphQL endpoint, MCP server, developer portal, or commerce surface.
• Emerging or bonus: experimental signals that may help but should not displace fundamentals.

↳

The report’s evidence is more useful than the headline number. A score can change when the website changes, the methodology changes, or a time-sensitive check returns a different result.

Step 03

Use the generated prompt as a reviewed implementation brief

The report’s “Prompt to fix” converts actionable findings into a website-specific implementation brief. Copy it into Cursor, Claude Code, Gemini CLI, GitHub Copilot, Windsurf, or another coding assistant only after reading the findings yourself.

Ask the coding assistant to preserve existing behavior, explain each proposed change, and work in small reviewable commits. A generated prompt can misunderstand architecture, recommend unnecessary files, or optimise for the scanner instead of users.

Fix the verified issues in this audit while preserving existing functionality, design, accessibility, analytics, security controls, and SEO behavior.

Before editing:
1. Inspect the existing project architecture and relevant configuration.
2. Match every proposed change to evidence in the audit.
3. Flag findings that appear incorrect, unsafe, or not applicable.
4. Explain the implementation plan and files that will change.
5. Ask before adding dependencies or making architectural changes.

After editing:
1. List every modified and created file.
2. Run the existing tests, lint, type checks, and production build.
3. Explain how to verify each fix on the deployed website.
4. Do not claim the changes guarantee AI citations, recommendations, rankings, or traffic.

↳

Treat scanned content and generated prompts as untrusted input. Never paste secrets, approve unrelated commands, or allow broad changes without reviewing the diff.

Step 04

Check the crawler policy you actually intend

Crawler names are not interchangeable. OpenAI documents OAI-SearchBot for ChatGPT search visibility separately from GPTBot for potential model-training use. Anthropic similarly documents separate bots for search, user-requested retrieval, and model development.

Review robots.txt, CDN bot rules, web application firewalls, and server logs. Allow or block each crawler according to your business and content policy; do not copy a blanket rule without understanding it.

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

↳

Crawler access changes eligibility, not selection. Allowing a search bot does not guarantee inclusion in an answer, and blocking a training bot does not necessarily block a separate search bot.

Step 05

Make the important answer available in the initial HTML

A public page should return a successful status and meaningful content without depending entirely on client-side JavaScript. Critical product information should not exist only inside images, popups, interactions, login walls, or components that load after an avoidable delay.

• Return the correct 2xx, 3xx, 4xx, and 5xx responses instead of soft-error pages.
• Server-render or statically generate the primary heading, explanation, and important links where practical.
• Use text for core claims and provide descriptive alternative text for meaningful images.
• Keep public documentation readable without an account when the content is intended for discovery.
• Test the production response as a crawler would receive it, not only the hydrated browser view.

Step 06

Use semantic structure and specific product language

Clear HTML helps machines and people understand page hierarchy. Use one descriptive H1, ordered H2 and H3 sections, semantic landmarks, real lists, descriptive links, a canonical URL, a declared language, and accurate metadata.

Replace broad claims with concrete explanations of the user, problem, input, action, and outcome. “We revolutionize workflows” carries little usable meaning. “We generate invoices from WhatsApp orders” gives a retrieval system something precise to match.

• Give every important page one clear search intent and audience.
• Explain what the product does before introducing brand language or technical architecture.
• Add accurate title, description, canonical, Open Graph, and relevant structured data.
• Avoid keyword stuffing; repeated vague phrases do not create understanding.

Step 07

Publish useful documentation and connect every important page

Public documentation can clarify a SaaS product, API, open-source project, or developer tool more precisely than a marketing homepage. Document real capabilities, inputs, outputs, limitations, pricing boundaries, examples, and common questions.

• Link important pages through normal HTML links rather than making them discoverable only through site search or scripts.
• Use descriptive anchor text that explains what the destination answers.
• Keep orphan pages out of production and update links when routes change.
• Maintain an XML sitemap and intentional robots policy as discovery aids, not substitutes for navigation.
• Publish dates or update dates when freshness materially affects the answer.

Step 08

Deploy, rescan, and measure outcomes separately

Review the diff, run the project’s checks, deploy to production, and scan the same canonical URL again. Is Agentic notes that reports are snapshots and may use cached results, so check the report timestamp before judging a deployment.

Track technical readiness separately from business results. A score shows what one scanner observed. Referral traffic, cited mentions, qualified sign-ups, branded searches, and conversions show whether discovery is improving for the business.

• Verify each changed response, page, metadata field, and crawler rule in production.
• Rescan and compare failed checks, evidence, timestamps, and score-not only the score.
• Monitor server logs for intended crawlers and unexpected access failures.
• Test representative questions in relevant assistants without treating one answer as a stable ranking.
• Repeat after framework migrations, redesigns, domain changes, documentation changes, or major releases.

Save this checklist

AI visibility audit checklist

• Scan the canonical production URL on Is Agentic.
• Read the evidence and fix Essential issues first.
• Review crawler permissions and CDN or firewall behavior.
• Ensure important content exists in meaningful server-rendered or static HTML.
• Improve semantic headings, metadata, specific copy, documentation, and internal links.
• Review the generated prompt before applying any code changes.
• Run tests and deploy the verified fixes.
• Rescan the same URL and confirm the report timestamp changed.
• Measure citations, referral traffic, sign-ups, and conversions independently.