2026 AI Visibility Benchmark: 67 Pages Analyzed

GEOCARA analyzed 67 public pages across 15 distinct domains with the current grader-readiness-v2 model between August 3 and August 14, 2026. The median GEO Authority Score was 40.3/100. Information density was flagged on every domain, while Answer First and multimodal weaknesses each affected 80% of the sample.

Executive summary

This first GEOCARA AI Visibility Readiness Benchmark measures whether websites expose the technical, structural, and editorial signals needed for answer engines to retrieve, understand, trust, and cite their content. It is a readiness benchmark, not a report of live mentions inside ChatGPT or other AI products.

The sample is deliberately modest and the findings should be treated as directional. Its value is transparency: the cohort, scoring version, date range, aggregation rules, and limitations are all stated below.

Benchmark measure	Result
Distinct domains	15
Public pages analyzed	67
Audit period	August 3-14, 2026
Scoring model	grader-readiness-v2
Mean GEO Authority Score	39.4/100
Median GEO Authority Score	40.3/100
Domains below 40	46.7%
Domains from 40 to 69.9	46.7%
Domains at 70 or above	6.7%

The current free AI visibility checker uses the same eight-category readiness framework. Its public result shows the overall score and selected priority categories without requiring an account.

Average score by readiness category

HTML structure was the strongest category in the sample, with a mean of 76.7. Section length was the weakest at 19.0, followed by multimodal content at 30.0 and Answer First at 30.9.

Readiness category	Mean score
HTML structure	76.7
E-E-A-T	46.0
Schema quality	41.5
Informational density	36.7
FAQ presence	34.7
Answer First	30.9
Multimodal content	30.0
Section length	19.0

These averages describe the GEOCARA model's findings, not universal search-engine thresholds. A higher category score indicates that more of the audited pages satisfied the criteria defined in the AI Visibility Checker methodology.

The most common recommended improvements

For each distinct domain, GEOCARA retained the latest completed V2 audit and counted whether each recommendation type appeared at least once. This prevents repeat scans of the same domain from inflating the totals.

Recommended improvement	Domains affected	Share of sample
Improve informational density	15 of 15	100.0%
Add or improve Answer First openings	12 of 15	80.0%
Add useful multimodal content	12 of 15	80.0%
Add or improve FAQ coverage	11 of 15	73.3%
Add or improve schema markup	11 of 15	73.3%
Strengthen content maturity and evidence	10 of 15	66.7%
Strengthen on-page E-E-A-T	10 of 15	66.7%
Improve section development	9 of 15	60.0%
Clarify brand identity signals	6 of 15	40.0%
Review freshness signals	4 of 15	26.7%

Finding 1: clear HTML is common, extractable answers are not

The average HTML structure score of 76.7 suggests that many audited sites already use a recognizable H1 and H2 hierarchy. That foundation did not translate into equally strong answers: the Answer First average was only 30.9, and 80% of domains received a direct-answer recommendation.

This distinction matters. A page can have valid headings while still opening with a slogan, a vague promise, or several paragraphs of context before answering the visitor's question. The practical fix is not to add more headings. It is to make the first substantive paragraph useful on its own, then use the rest of the page to qualify and support it.

Finding 2: thin evidence is the only universal gap in this sample

Every V2 domain received an informational-density recommendation, and the average density score was 36.7. The model looks for concrete facts, examples, named entities, and substantive detail rather than raw word count.

That does not mean every page should be longer. Google's own SEO guidance says there is no magical minimum or maximum content length. Useful content is original, well organized, current, reliable, and written for people (Google Search Central). A concise pricing page with exact limits can be denser than a 2,000-word article built from generic claims.

Finding 3: schema adoption remains incomplete

Schema quality averaged 41.5, and 73.3% of domains received a schema recommendation. Common gaps include a missing primary page entity, isolated JSON-LD blocks that are not connected through stable identifiers, or FAQ markup that does not match visible page content.

Structured data helps a crawler identify what a page represents and how it relates to the publisher. It does not replace the visible content and does not guarantee a rich result or AI citation. Google explicitly requires structured data to represent visible, relevant content and warns against misleading markup (Google Search Central).

Finding 4: trust is broader than adding an author name

The average E-E-A-T score was 46.0. Two-thirds of the sample received recommendations related to content maturity and on-page responsibility. In the current model, E-E-A-T combines four layers: authorship and dates, domain trust, brand identity pages, and content evidence.

This explains why adding one byline rarely closes the gap. Visitors and retrieval systems also need to know who operates the website, how to contact the organization, when information was reviewed, and which sources support important claims.

Finding 5: useful visuals are underused

Multimodal content averaged 30.0, and 80% of domains received a corresponding recommendation. The objective is not decorative imagery. Screenshots, diagrams, tables, charts, and short demonstrations can make a claim easier to verify and a process easier to understand.

Images should sit near the text they explain and include descriptive alternative text. Google recommends using clear, high-quality images near relevant content so search systems and users can understand the connection (Google Search Central).

Methodology

The dataset was generated from completed public GEOCARA grader audits stored between August 3 and August 14, 2026.

1. Only rows produced by grader-readiness-v2 with a completed status and a non-null global score were eligible.
2. The latest completed audit was retained for each normalized domain.
3. Domain names, email addresses, IP addresses, and individual report identifiers were excluded from the analysis.
4. Site-level means were calculated from the stored sub-scores of the 15 retained audits.
5. Recommendation frequency counts each domain at most once per recommendation type.
6. The 67-page total is the sum of pages successfully analyzed in the retained audits.

The global score is the mean of eight sub-scores. The full scoring definitions and limitations are published in the GEOCARA checker methodology.

Limitations

• Small sample: Fifteen domains are not representative of the entire web or a specific industry.
• Self-selection: Website owners chose to run the public checker, so the cohort is not randomly sampled.
• Five-page maximum: Each audit covers up to five pages and may miss weaknesses elsewhere on the domain.
• Readiness, not observed visibility: The benchmark does not measure live prompt mentions, citation share, sentiment, traffic, or conversions.
• No sector weighting: Results are not adjusted for industry, geography, company size, or website type.
• Model-specific results: Scores are tied to grader-readiness-v2 and should not be compared directly with a different scoring version.

GEOCARA will expand the cohort before publishing stronger sector conclusions. Future editions should preserve versioned cohorts and publish any scoring changes alongside the results.

What website teams should do first

The data suggests a practical sequence:

1. Rewrite the opening of each cornerstone page as a direct, self-contained answer.
2. Replace generic claims with specific facts, examples, limitations, and cited evidence.
3. Validate visible content and connected JSON-LD together.
4. Add responsibility signals: author or editorial owner, review date, organization identity, and contact path.
5. Add useful screenshots, tables, or diagrams where they clarify the answer.
6. Re-run the same audit model after the changes and compare the underlying categories, not only the total score.

For a step-by-step measurement workflow, read how to check AI visibility.

FAQ

What is the average AI visibility readiness score in this benchmark?

The mean GEO Authority Score was 39.4/100 and the median was 40.3/100 across 15 distinct domains analyzed with grader-readiness-v2. These figures are model-specific readiness scores, not universal industry averages.

What was the most common AI-readiness problem?

All 15 domains received an informational-density recommendation. Answer First and multimodal recommendations were next, each appearing for 12 of the 15 domains.

Does this benchmark show which sites are mentioned by ChatGPT?

No. It measures technical and content readiness across a small page sample. Live mention tracking requires recurring prompts submitted to specific engines and must be reported separately.

Can these results be compared with another vendor's score?

Not directly. Vendors use different crawlers, samples, definitions, and weights. Compare the measured components and methodology before comparing headline numbers.

Will GEOCARA update this research?

Yes. Later editions can expand the number of domains and introduce sector cohorts once each cohort is large enough to support responsible comparisons.

Sources

• GEOCARA: AI Visibility Checker methodology
• GEOCARA: Free AI Visibility Checker
• Google Search Central: SEO Starter Guide
• Google Search Central: Structured data general guidelines