We Audited 98 Websites for AI Search Readiness. Here's What We Found.

I ran 100 audits with my tool. Here's what the data actually looks like.

I built getaisearchscore.com to measure how well websites are set up for AI search citation. Between January and March 2026, 98 real websites went through the full audit pipeline: 26 automated checks, scoring across four dimensions, and 1,615 citation checks against Perplexity. The average score was 52.8 out of 100. Only 18.1% of AI search queries cited the audited sites. And 91% of websites had zero customer review markup.

Below is everything I found. No cherry-picking, no marketing spin. Just the numbers and what I think they mean.

How I Collected This Data

Every scan came from a real user who submitted their site to getaisearchscore.com. I did not curate or hand-pick the sample. Each scan runs 26 automated checks across four dimensions:

• Machine Readability (MR) — can AI crawlers access and parse your content?
• Extractability (EX) — is your content structured so AI can extract citable answers?
• Trust & Entity (TR) — does your site demonstrate authoritativeness and verifiable identity?
• Offering Readiness (OR) — for e-commerce: are products structured for AI shopping?

After scoring, each site got 20 AI-generated monitoring queries sent to the Perplexity API to check actual citation rates. Total: 1,615 citation checks across all 98 sites.

The sample is self-selected. People who find and use an AI readiness tool are already more aware than the average site owner. Real-world numbers are probably worse.

The Numbers at a Glance

The average of 52.8 sounds mediocre, and it is. But remember this is a self-selected group of people who care enough to audit their sites. The real average across the web is lower.

Score Distribution

Most sites cluster in the 20–59 range. There is a surprisingly strong cohort above 80. The 60–79 middle ground is the thinnest segment.

The shape is bimodal. Sites either have fundamental issues (20–59) or have already implemented the key signals (80+). The thin middle at 60–79 suggests that once you fix the core problems, the score jumps. There is no gradual climb.

The 10 Most Common Failures

I ranked all 26 checks by the percentage of sites that scored zero — not partially implemented, but completely absent.

What I Take Away from This

The top 3 failures are about trust and answer-readiness, not technical SEO. 91% of sites have zero review markup. Two-thirds lack FAQ content that AI could extract as direct answers. 60% show no authorship signals at all.

Most websites have already solved the technical accessibility basics. They fail on the signals that are supposed to help AI engines select citation sources. Whether AI engines actually use these signals for selection is a separate question — my follow-up research suggests the relationship is weaker than the industry assumes.

What Almost Everyone Gets Right

Not everything is broken. Several checks had near-universal pass rates:

Every check here is a Machine Readability or basic Extractability signal. These are the foundations of traditional SEO: HTTPS, proper headings, indexable content. The 90%+ pass rate means this is no longer a differentiator. Everybody has it.

The Trust Gap

The clearest pattern in the data is the disconnect between technical SEO quality and trust signals. A site can have perfect SSL, clean robots.txt, fast mobile performance, proper headings — and still score below 40/100 because trust and entity signals are missing.

The three trust checks (TR3, TR5, TR6) have a combined zero-score rate of 72%. The six Machine Readability checks average just 11%. That is a 6x gap.

• Review markup (AggregateRating, Review schema) is almost universally missing
• Authorship signals (bylines, author bio, credentials) are absent on 60% of content pages
• Product identifiers (GTIN, MPN, SKU) are missing on 66% of e-commerce sites

Sites invest in crawlability but neglect verifiable identity. That said — I want to be careful about implying causation here. My later research found that fixing these gaps does not reliably increase citations. The gap is real; whether closing it matters for AI visibility is less certain.

Citation Rate: 18.1%

After each scan, I generate 20 monitoring queries relevant to the site's content and niche, then check whether Perplexity cites the site. Across 1,615 total citation checks:

• 292 citations found (18.1% citation rate)
• 1,323 queries where the site was not cited despite being topically relevant

An 18.1% citation rate means that even for queries directly related to a site's products or services, Perplexity chose other sources more than 4 out of 5 times. Competition for citation slots is fierce.

I initially assumed that higher-scoring sites would get cited more. They don't. When I later tested this on 441 domains, the correlation was essentially zero (r=0.009). The factor that actually predicted citations was content relevance — whether the page directly addressed the query topic.

The Schema.org Paradox

100% of crawled pages had some form of Schema.org structured data. Yet 36% scored zero on the Schema.org check (MR3). How?

Because having some schema is not the same as having useful schema. Most sites include basic Organization or WebSite types but lack the specific structured data that matters for citation decisions:

• Product + Offer schemas with price, availability, and identifiers (GTIN/MPN)
• AggregateRating with reviewCount and ratingValue
• FAQPage with question-answer pairs
• Article + Author with byline linking to a real person

Schema.org presence is table stakes. Schema.org depth is what separates a score of 28 from a score of 67. In my data, sites with rich schema scored 38 points higher on average (66.7 vs 28.7).

Score by Business Vertical

The sample is too small for statistically significant comparisons. But the pattern is visible:

Niche e-commerce sites (dive shops, candle makers, pet food brands) scored 12 points lower. They tend to have deep product knowledge but fail to express it in structured formats. This is the most fixable gap I see — the content exists, it just is not machine-readable.

What I Would Tell a Site Owner Based on This Data

1. Add Review Markup

91% of sites have zero review markup. If you have customer reviews, add AggregateRating and Review Schema.org. If you do not have reviews, start collecting them. This is the most common gap in the dataset.

2. Create FAQ Content

66% of sites have no extractable FAQ content. Add an FAQ section to key pages with concise answers (2–3 sentences each). Use FAQPage schema. AI engines are question-answering machines — give them questions and answers.

3. Add Authorship Signals

60% of sites show no author identity. Add bylines, bios, credentials, links to professional profiles. This costs nothing and addresses a widespread gap.

4. Lead with the Answer

52% scored zero on content clarity. Put your main answer in the first paragraph. AI extracts the first semantically complete passage. If you bury the answer, someone else's content gets extracted instead.

5. Do Not Confuse SEO with AI Readiness

90%+ pass traditional SEO checks. 50–91% fail AI-specific checks. If your SEO tool says everything is fine but AI is not citing you, the tool is not measuring the right things.

6. But Manage Your Expectations

I built this tool assuming that better-structured sites get cited more. My own data from 441 domains says that is not the case. Content relevance — whether your page directly addresses the question being asked — is the factor that actually matters (62x difference). Structural optimization is good practice, but it is not a citation guarantee.

Limitations

I want to be upfront about what this data can and cannot tell you:

• Self-selected sample — users who found my tool are more AI-aware than average. Real failure rates are likely higher.
• Sample size — 98 scans shows patterns but is not enough for statistically significant vertical or geographic breakdowns.
• Perplexity only — I track citations on Perplexity. ChatGPT and Google AI Overviews may behave differently.
• Temporal snapshot — this data is from Jan–Mar 2026. AI search changes fast.
• Score ≠ citations — my follow-up study (441 domains, 14,550 pairs) found r=0.009 between score and citation rate. These readiness metrics describe site structure, not citation likelihood.

If you want your site in this dataset, you can run a free scan at getaisearchscore.com. The tool still works and the audit is still useful for understanding your site's structure. Just do not expect that a higher score automatically means more AI citations.

Full Methodology

Each scan runs 26 automated checks: 9 core (visible to all users) and 17 additional (premium + LLM-based). The scoring formula:

Each basket is scored as a percentage of its maximum possible points. Extractability gets the highest weight (30%). A JavaScript rendering penalty halves the MR subscore if the static HTML contains fewer than 50 words.

Citation monitoring sends 20 queries per site to the Perplexity API. Query generation uses GPT-4o to create realistic search queries based on the site's actual content and niche.

The full scoring methodology is documented here. The follow-up correlation study is documented here.