How to Check AI Visibility: A 3-Layer Guide

To check AI visibility accurately, measure three separate layers: whether AI search crawlers can retrieve and understand your pages, whether your brand is actually mentioned or cited for a fixed set of prompts, and whether those answers send qualified visits or conversions. No single scan can prove all three outcomes.

The three layers of AI visibility

“AI visibility” is often used for several different measurements. That makes reports sound comparable when they are not.

Layer	Question it answers	Evidence
Readiness	Can engines retrieve, understand, and trust the website?	Crawl access, content structure, schema, evidence, entity signals
Observed visibility	Does the brand appear for relevant questions?	Repeated prompt results, mentions, citations, position, competitors
Business impact	Did AI visibility produce useful traffic or outcomes?	Referrals, engagement, sign-ups, assisted conversions, revenue

Start with readiness because blocked or ambiguous pages limit everything downstream. Then test prompts because a technically excellent website can still be absent from real answers. Finally, connect citations and referrals to business outcomes so visibility does not become a vanity metric.

Step 1: establish a readiness baseline

Run a technical and content scan of the pages that matter most: homepage, product or service pages, pricing, about, and high-value guides. The scan should check at least:

• successful public HTTP access and indexability;
• robots.txt rules for the search crawlers you intentionally allow;
• one clear H1 and a descriptive section hierarchy;
• a direct answer near the beginning of the page;
• valid, visible-content-aligned structured data;
• author, date, organization, and contact signals;
• factual evidence and external sources;
• useful images, tables, or demonstrations.

GEOCARA's free AI visibility checker scans up to five pages and produces a readiness baseline without sign-up. The exact scoring rules are documented in the checker methodology.

Record the date, scoring-model version, pages analyzed, overall score, and each sub-score. A headline score without the underlying categories is difficult to act on and risky to compare over time.

Step 2: verify search and crawler access manually

Open the following resources in a clean browser session:

1. https://yourdomain.com/robots.txt
2. https://yourdomain.com/sitemap.xml
3. the canonical URL of each cornerstone page
4. the rendered HTML or URL Inspection result for those pages

Check that the page returns a successful response, has no accidental noindex, uses the preferred canonical URL, and exposes its important text without requiring authentication or a challenge page.

Crawler names have different purposes. For ChatGPT search inclusion, OpenAI currently tells publishers not to block OAI-SearchBot; GPTBot controls potential model-training access and is a separate policy decision (OpenAI Publisher FAQ). For Google AI features, ordinary Google Search eligibility remains the foundation (Google Search Central).

Step 3: build a fixed prompt set

Readiness does not prove that a brand appears in answers. To measure observed visibility, create 15-30 prompts based on real buyer questions and keep them stable for several measurement cycles.

Include five prompt types:

• Category: “What is the best software for [job]?”
• Problem: “How can a team solve [specific problem]?”
• Comparison: “[Competitor] alternatives for [audience]”
• Evaluation: “Which [category] tools support [requirement]?”
• Brand: “What is [brand], and who is it for?”

Favor unbranded questions for discovery measurement. A branded prompt proves that an engine recognizes the name; it does not prove that the brand wins when a buyer has not chosen a vendor.

Step 4: test the same prompts across engines

Run the fixed set through the engines your audience uses. At minimum, consider ChatGPT search, Google AI features, Perplexity, Gemini, Claude web search, and Microsoft Copilot where available in the target market.

For Microsoft's ecosystem, complement manual prompts with the first-party Bing AI Performance report, which separates cited pages and sampled grounding queries from ordinary Bing Search clicks.

For every prompt-engine combination, record:

• date and market or language;
• whether the brand was mentioned;
• whether a GEOCARA-owned URL was cited;
• the exact cited URL;
• position or prominence in the answer;
• competing brands and cited domains;
• a saved answer or screenshot;
• sentiment or factual accuracy.

Use fresh sessions where practical and run more than one observation before treating a change as a trend. Generative answers can vary between runs even when the prompt is identical.

Step 5: calculate interpretable metrics

Keep mentions and citations separate:

• Mention rate = prompts with a brand mention / prompts tested
• Citation rate = prompts with a link to the brand's domain / prompts tested
• Prompt coverage = prompts where the brand is mentioned or cited / total tracked prompts
• Share of voice = brand mentions / mentions of all tracked brands

Report each metric by engine, market, and prompt group. Combining every engine into one number can hide a serious gap, such as strong visibility in Perplexity but no presence in ChatGPT.

The detailed AI mention-tracking guide includes a spreadsheet-ready field list and cadence recommendations.

Step 6: measure AI referral traffic

Use an analytics platform to create a dedicated AI-referral channel group. Inspect referrers and campaign parameters from ChatGPT, Perplexity, Gemini, Copilot, Claude, and other relevant surfaces. OpenAI states that ChatGPT referral URLs include utm_source=chatgpt.com, which can support attribution in analytics tools (OpenAI Publisher FAQ).

The step-by-step ChatGPT referral traffic guide shows how to validate this source in GA4 and Plausible, connect it to landing pages, and measure successful leads or registrations without counting bots as visitors.

Track more than sessions:

• engaged sessions and time on useful pages;
• checker completions or product interactions;
• account creation;
• qualified leads;
• assisted conversions;
• revenue when applicable.

A citation that produces no visits may still influence awareness. A small number of high-intent referrals may be commercially more valuable than a large volume of informational impressions. Keep both observations visible.

Step 7: connect findings to specific changes

Each visibility gap should create a testable action:

Observation	Likely next action
Crawler blocked	Correct robots.txt, CDN, firewall, or challenge rules
Page retrieved but never cited	Improve direct answers, evidence, entity clarity, and topical relevance
Competitor cited instead	Compare the exact competing source's structure, facts, and freshness
Mention exists but is inaccurate	Publish and distribute a canonical, sourced correction
Citation exists but produces no conversion	Improve message match, landing-page clarity, and next action

Log the publication or technical-change date beside your measurement series. Without an intervention timeline, a later movement cannot be attributed responsibly.

A practical reporting cadence

• Weekly: high-priority prompts during launches or active optimization.
• Monthly: stable prompt set, engine-level mention and citation trends.
• Quarterly: technical readiness audit and page-level content review.
• After major changes: recheck affected pages once crawlers have had time to refresh them.

Google notes that some search changes can appear within hours while others can take months, and generally recommends waiting several weeks before evaluating their impact (Google Search Central).

FAQ

Can I check AI visibility for free?

Yes. Use a free readiness checker for the technical baseline, then manually run a small fixed prompt set across the AI engines that matter to your audience. Analytics and search-console data can provide the traffic layer.

Is an AI visibility score enough?

No. A readiness score diagnoses whether pages expose useful signals. It does not prove live mentions, citations, traffic, or conversions. Those must be measured separately.

How many prompts should I track?

Start with 15-30 stable, commercially relevant prompts. A smaller representative set measured consistently is more useful than hundreds of changing prompts that cannot be compared over time.

How often should prompts be tested?

Monthly is a reasonable baseline. Use weekly checks during active campaigns, product launches, or concentrated content work, and preserve the same prompts long enough to distinguish movement from noise.

Which AI visibility metric matters most?

It depends on the decision. Readiness helps prioritize website fixes, citations prove sourced presence, share of voice supports competitive analysis, and qualified referrals or conversions show business impact.

Sources

• OpenAI: Publishers and Developers FAQ
• Google Search Central: AI features and your website
• Google Search Central: SEO Starter Guide
• GEOCARA: AI Visibility Checker methodology
• GEOCARA: 2026 AI Visibility Readiness Benchmark