No. Traditional SEO optimises for Google's keyword-matching algorithm. AI search readiness focuses on whether AI systems (ChatGPT, Perplexity, Claude) can understand, trust, and cite your content when answering user questions. The signals overlap somewhat (good structure, clear content, schema markup) but AI search adds distinct requirements: explicit bot permissions in robots.txt, llms.txt, entity disambiguation, answer-ready content format, and E-E-A-T depth. A site can rank well in Google and still be nearly invisible to AI-powered answers.

Why does my score change between scans?

Several reasons. First, your actual content may have changed, PromptScore detects this via a content hash. Second, AI-graded checks (homepage clarity, query coverage) are cached for 30 days tied to both your content hash and the current scoring version; if either changes, the check re-runs. Third, external signals like PageSpeed Insights vary based on real-world server performance at the time of scan. If you see an unexpected score drop, compare the two scan results' evidence fields to identify what changed.

Why are some checks graded by AI?

Purely deterministic checks can only measure presence or absence. Whether your homepage clearly explains what you do, who it is for, and what outcome you deliver is a nuanced judgement that no regex can make reliably. We use Claude at temperature 0 with a locked, version-controlled prompt and a strict JSON output schema. Temperature 0 means the model is essentially deterministic. All AI-graded results are cached against the content hash, so rescanning the same unchanged page returns the identical score.

What is the difference between retrieval bots and training bots?

Retrieval bots (OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBot) fetch your content at the time a user asks a question. Allowing them means your site appears in live AI-powered answers. Training bots (GPTBot, ClaudeBot, CCBot, Google-Extended) crawl your content to train the underlying model. Allowing them means your content may influence the model's base knowledge. These are separate decisions with separate business implications. PromptScore scores whether you have an explicit stance on each group, not which choice you make.

We block PerplexityBot via Cloudflare. Should we unblock it?

Almost certainly yes, if appearing in Perplexity answers is valuable to you. PerplexityBot is a retrieval bot: it fetches your content at query time, not to train a model. Blocking it means Perplexity cannot include your site in answers, regardless of how good your content is. The exception would be if you have a specific legal or commercial reason to prevent Perplexity from citing you. To unblock: Cloudflare dashboard, Security, Bots, allow PerplexityBot and Perplexity-User user agents. Also check your robots.txt does not disallow them.

How often should we re-scan?

Scan after any significant change to your site: new schema markup, updated robots.txt, content rewrites, technical migrations. For ongoing monitoring without changes, once per month is sufficient. AI search ranking signals change more slowly than keyword rankings, weekly scanning rarely surfaces meaningful changes. If you are running a structured improvement programme, scan at the start and end of each implementation sprint.

Can I get a private scan that is not visible to others?

Scan results are not publicly listed or indexed. A scan result is accessible only to someone who has the direct URL containing the scan ID (a non-guessable UUID). There is no public directory of scans. If you need a result removed entirely, contact us at lee@performancepeak.co.uk and we will delete it from the database.

Who built PromptScore?

PromptScore is built and operated by Performance Peak, a digital marketing and technology consultancy based in the UK. The scoring model draws on public AI crawler documentation from OpenAI, Anthropic, Google, and Perplexity, as well as the emerging llmstxt.org specification and the Schema.org vocabulary. The methodology is published here so it can be scrutinised and improved.

How can I improve my score?

The full PDF report (available after entering your email on the results page) contains the top five priority actions ranked by impact divided by implementation effort. Start there. In general: fix your robots.txt first (it gates everything), add or complete your Organization schema, ensure your homepage clearly states what you do and who for in the first viewport, and add llms.txt. Most sites can move from Significant gaps to Solid foundation with two to four weeks of focused technical work.

Is the score guaranteed accurate?

No. The score is a systematic estimate based on observable signals, not a direct measurement of AI citation frequency. It does not guarantee that a high-scoring site will appear in any specific AI answer, or that a low-scoring site will not. AI search ranking is influenced by many factors PromptScore does not measure, including the AI provider's own content policies, query intent, and real-time retrieval decisions. The value of the score is its repeatability: the same inputs reliably produce the same score, so you can measure progress.

How PromptScore works

Every check, every weight, and how we calculate your score. Published here so you can verify, critique, and improve against it.

Written by Lee Debnam, founder of Performance Peak. Published 15 April 2026. Last updated 5 May 2026.

Scoring model version: 1.0.0

References and source documents

The check list and weights are derived from public specifications and crawler documentation: Schema.org, llmstxt.org, OpenAI bots documentation, Anthropic ClaudeBot documentation, Google Search Central crawler index, and the W3C TDM Reservation Protocol.

The five categories

Your overall score is a weighted sum across five categories. The weights reflect the relative importance of each category to AI search discoverability as of the current scoring version.

Category	Weight	What it measures
Crawler access	20%	Whether AI search bots (retrieval and training) can reach and render your content. Covers robots.txt, sitemap, HTTPS, WAF posture, JS-rendering dependency.
Structured data	25%	Schema.org markup (JSON-LD), Open Graph, Twitter cards, and entity signals that let AI parse what the business is and what each page is about.
Content clarity	25%	Semantic HTML, heading hierarchy, FAQ patterns, answer-ready content structure, publication dates, author attribution. Partly deterministic, partly AI-scored.
AI-specific signals	15%	llms.txt, AI-crawler-specific directives, explicit handling of retrieval bots vs training bots, TDM reservation headers.
Authority & trust	15%	E-E-A-T signals: author pages, About page depth, contact info, citations and sources, Wikipedia/Wikidata presence, brand consistency.

How checks are run

Each check is one of three types. Every check returns a score of 0, 0.5, or 1 (or a continuous value for composite checks, normalised before aggregation).

DDeterministic

Pure parsing or HTTP check. Returns the same answer every time.

DCComposite

Combines several deterministic signals using fixed thresholds.

AAI-graded

Uses Claude at temperature 0 with a locked prompt. Cached by URL, content hash, and prompt version.

Approximately 75% of the overall score comes from D and DC checks. AI-graded checks (type A) are used only within Content clarity and Authority & trust, where human-readable judgement is required.

All 36 checks

This table is generated directly from the scoring package at build time. Adding or changing a check in code automatically updates this page.

Crawler access

20% weight · 7 checks

Check	Type	Weight	What it checks
robots.txt is present and valid robots_valid	D	2	A well-formed robots.txt gives AI crawlers clear permission signals from the first request.
AI retrieval bots can access your site retrieval_bots_allowed	D	5	OAI-SearchBot, PerplexityBot, and Claude-SearchBot can crawl and index your content for AI search results.
Explicit stance on AI training bots training_bots_explicit	D	3	You've made a clear decision about GPTBot, ClaudeBot, and Google-Extended, AI providers respect explicit rules.
Sitemap linked from robots.txt sitemap_present_linked	D	2	AI crawlers find your full content inventory via the Sitemap directive in robots.txt.
HTTPS with HSTS enabled https_hsts	D	2	HSTS tells browsers and crawlers to always use encrypted connections, a trust signal that AI systems factor in.
Content renders without JavaScript js_dependency_ratio	DC	5	Most of your content is visible in the raw HTML, AI crawlers that don't execute JS can read everything.
Fast mobile performance pagespeed_mobile	DC	1	A PageSpeed score ≥75 signals well-structured, lightweight pages that crawlers can process quickly.

Structured data

25% weight · 8 checks

Check	Type	Weight	What it checks
Organization schema markup schema_organization	D	5	JSON-LD Organization schema tells AI exactly who you are, what you do, and how to find you.
Schema types match your business category schema_category_appropriate	DC	6	Your schema types (Hotel, LegalService, Product, etc.) match what AI systems expect for your category.
FAQ/HowTo schema on Q&A content schema_faq_howto	DC	3	FAQ schema on your Q&A content makes it directly quotable in AI-generated answers and featured snippets.
BreadcrumbList schema on inner pages schema_breadcrumbs	D	2	Breadcrumb schema helps AI understand your site's information hierarchy and which pages are most important.
Complete Open Graph tags og_tags_complete	D	3	Full OG tags (title, description, image, url, type) ensure consistent previews across AI-powered social and search surfaces.
Twitter/X card with image twitter_card	D	2	Twitter card metadata ensures your content displays with an image when shared or cited in AI-generated content.
Schema markup has no errors schema_validates	D	2	Your structured data passes validation, no parsing errors that could cause AI systems to silently discard it.
Canonical URLs set correctly canonical_urls	D	2	Self-referential canonical tags prevent duplicate content confusion for AI crawlers.

Content clarity

25% weight · 8 checks

Check	Type	Weight	What it checks
Logical heading structure (H1–H6) heading_hierarchy	D	3	Clean heading hierarchy lets AI parse your content structure and identify the most important topics on each page.
Semantic HTML landmarks semantic_landmarks	D	2	Proper use of <main>, <nav>, <header>, <footer>, and <article> lets AI skip boilerplate and focus on your core content.
Images have descriptive alt text alt_text_coverage	DC	2	Descriptive alt text on images extends your content into AI's understanding of your visual assets.
Publication and update dates marked up publication_dates	DC	2	Dated content with datePublished and dateModified schema helps AI systems surface your freshest, most relevant pages.
Homepage clearly explains what you do homepage_clarity_rubric	A	6	Your homepage immediately communicates what you do, who it's for, and the outcome, perfect for AI citation.
Content answers real user queries query_coverage_rubric	A	6	Your content directly addresses the questions users ask AI assistants about your category, increasing citation frequency.
FAQ content detected faq_content_present	D	2	FAQ sections are frequently cited verbatim in AI answers, a direct pipeline from your site to AI responses.
Content depth and internal linking content_depth_linking	DC	2	Pages with substantial content (500+ words) and strong internal links signal topic authority to AI ranking systems.

AI-specific signals

15% weight · 6 checks

Check	Type	Weight	What it checks
llms.txt file present and valid llms_txt_present	D	3	A valid llms.txt signals proactive AI-readiness and gives LLM providers a curated summary of your content.
llms-full.txt extended version present llms_full_txt	D	1	llms-full.txt provides AI systems with richer context for deep queries about your business.
Retrieval and training bots treated differently retrieval_vs_training_differentiated	D	4	Your robots.txt distinguishes between bots that answer queries (allow) and bots that train models (your choice), the correct modern posture.
Firewall not blocking AI retrieval bots waf_not_blocking_ai_bots	DC	4	AI retrieval bots can reach your site from their data centers, no WAF or rate-limit interference.
AI policy page present ai_policy_page	D	2	A dedicated AI policy page shows AI providers you've thought carefully about your content's use in AI systems.
TDM / AI training reservation headers set tdm_headers	D	1	X-Robots-Tag with TDM reservation clearly signals your training data preferences to AI providers.

Authority & trust

15% weight · 7 checks

Check	Type	Weight	What it checks
Substantive About page (300+ words) about_page_substantive	DC	2	A detailed About page gives AI systems accurate background to represent your team, history, and mission.
Complete contact information in HTML and schema contact_info_complete	D	2	Full contact details (address, phone, email) in HTML and schema confirm your business is real and locatable.
Author bylines with bios on content author_bylines	DC	2	Named authors with bios are a key E-E-A-T signal, AI systems use authorship to assess content credibility.
Entity on Wikidata wikidata_presence	D	3	A Wikidata entry anchors your brand as a known entity, AI systems use this to disambiguate and validate your business.
3+ authoritative sameAs links sameas_links	D	2	Multiple sameAs links (LinkedIn, Crunchbase, Companies House) help AI systems confirm your identity across the web.
Content cites sources and links to authorities citation_practice	A	2	Citing sources and linking to authoritative references signals trustworthy, well-researched content to AI systems.
Consistent brand name across schema, titles, and copy brand_consistency	D	2	Your brand name is consistent across structured data, page titles, and visible copy, reducing AI disambiguation errors.

AI-graded checks

3 checks are graded by Claude rather than deterministic code. These are checks where a nuanced judgement is needed that no regex or threshold can reliably make.

Model and temperature

Claude Sonnet is used at temperature 0. Temperature 0 makes the model effectively deterministic, the same input reliably produces the same output.

Prompt versioning

Each AI check has a locked prompt ID. Changing the prompt increments the version, which invalidates the cache for that check across all stored scans for fairness.

Result caching

AI check results are cached by (check key, prompt version, content hash). If your page content has not changed, a re-scan returns the identical cached result within 30 days.

Failure handling

If Claude returns invalid JSON, one retry is attempted. If the retry also fails, the check is marked not_scored and excluded from aggregation. The remaining weights are renormalised so your score is not penalised for infrastructure issues.

Output schema

Every AI check uses a strict JSON schema for its output. The model is instructed to produce only the specified fields. This makes the output machine-readable and auditable.

AI-graded checks in this version:

AHomepage clearly explains what you do(content clarity)
AContent answers real user queries(content clarity)
AContent cites sources and links to authorities(authority trust)

Score bands

Score bands give context to the number. They are fixed thresholds, not percentile rankings, so your band does not change based on how other sites score.

85–100

AI-Ready Leader

Your site is in the top tier for AI discoverability. Focus on maintaining and extending your advantage.

70–84

Solid foundation

The fundamentals are in place. Targeted improvements will push you ahead of competitors.

55–69

Partial readiness

Several gaps are costing you AI visibility. A focused 2–4 week programme could lift you into the top tier.

35–54

Significant gaps

Your site is missing structural elements that AI systems rely on. Acting now avoids falling further behind.

0–34

High risk

Your site is effectively invisible to the current generation of AI retrieval systems. This is a strategic problem that compounds monthly.

Scoring version

Every stored scan records the scoring version it was run under. When weights, rubrics, or score bands change, the version increments and all AI-check caches are invalidated for fairness. You can always see which version produced a scan on the results page.

Version	Date	Changes
1.0.0(current)	15 Apr 2026	Initial public release. 34 checks across 5 categories (crawler access 20, structured data 25, content clarity 25, AI-specific 15, authority and trust 15). Three check types: deterministic (D), composite (DC), and AI-graded (A) using Claude at temperature 0 with locked prompt versions. Five score bands: AI-Ready Leader (85+), Solid foundation (70-84), Partial readiness (55-69), Significant gaps (35-54), High risk (0-34).

Version

Date

Changes

1.0.0(current)

15 Apr 2026

Initial public release.

34 checks across 5 categories (crawler access 20, structured data 25, content clarity 25, AI-specific 15, authority and trust 15).
Three check types: deterministic (D), composite (DC), and AI-graded (A) using Claude at temperature 0 with locked prompt versions.
Five score bands: AI-Ready Leader (85+), Solid foundation (70-84), Partial readiness (55-69), Significant gaps (35-54), High risk (0-34).

Frequently asked questions

Is this just SEO?: No. Traditional SEO optimises for Google's keyword-matching algorithm. AI search readiness focuses on whether AI systems (ChatGPT, Perplexity, Claude) can understand, trust, and cite your content when answering user questions. The signals overlap somewhat (good structure, clear content, schema markup) but AI search adds distinct requirements: explicit bot permissions in robots.txt, llms.txt, entity disambiguation, answer-ready content format, and E-E-A-T depth. A site can rank well in Google and still be nearly invisible to AI-powered answers.
Why does my score change between scans?: Several reasons. First, your actual content may have changed, PromptScore detects this via a content hash. Second, AI-graded checks (homepage clarity, query coverage) are cached for 30 days tied to both your content hash and the current scoring version; if either changes, the check re-runs. Third, external signals like PageSpeed Insights vary based on real-world server performance at the time of scan. If you see an unexpected score drop, compare the two scan results' evidence fields to identify what changed.
Why are some checks graded by AI?: Purely deterministic checks can only measure presence or absence. Whether your homepage clearly explains what you do, who it is for, and what outcome you deliver is a nuanced judgement that no regex can make reliably. We use Claude at temperature 0 with a locked, version-controlled prompt and a strict JSON output schema. Temperature 0 means the model is essentially deterministic. All AI-graded results are cached against the content hash, so rescanning the same unchanged page returns the identical score.
What is the difference between retrieval bots and training bots?: Retrieval bots (OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBot) fetch your content at the time a user asks a question. Allowing them means your site appears in live AI-powered answers. Training bots (GPTBot, ClaudeBot, CCBot, Google-Extended) crawl your content to train the underlying model. Allowing them means your content may influence the model's base knowledge. These are separate decisions with separate business implications. PromptScore scores whether you have an explicit stance on each group, not which choice you make.
We block PerplexityBot via Cloudflare. Should we unblock it?: Almost certainly yes, if appearing in Perplexity answers is valuable to you. PerplexityBot is a retrieval bot: it fetches your content at query time, not to train a model. Blocking it means Perplexity cannot include your site in answers, regardless of how good your content is. The exception would be if you have a specific legal or commercial reason to prevent Perplexity from citing you. To unblock: Cloudflare dashboard, Security, Bots, allow PerplexityBot and Perplexity-User user agents. Also check your robots.txt does not disallow them.
How often should we re-scan?: Scan after any significant change to your site: new schema markup, updated robots.txt, content rewrites, technical migrations. For ongoing monitoring without changes, once per month is sufficient. AI search ranking signals change more slowly than keyword rankings, weekly scanning rarely surfaces meaningful changes. If you are running a structured improvement programme, scan at the start and end of each implementation sprint.
Can I get a private scan that is not visible to others?: Scan results are not publicly listed or indexed. A scan result is accessible only to someone who has the direct URL containing the scan ID (a non-guessable UUID). There is no public directory of scans. If you need a result removed entirely, contact us at lee@performancepeak.co.uk and we will delete it from the database.
Who built PromptScore?: PromptScore is built and operated by Performance Peak, a digital marketing and technology consultancy based in the UK. The scoring model draws on public AI crawler documentation from OpenAI, Anthropic, Google, and Perplexity, as well as the emerging llmstxt.org specification and the Schema.org vocabulary. The methodology is published here so it can be scrutinised and improved.
How can I improve my score?: The full PDF report (available after entering your email on the results page) contains the top five priority actions ranked by impact divided by implementation effort. Start there. In general: fix your robots.txt first (it gates everything), add or complete your Organization schema, ensure your homepage clearly states what you do and who for in the first viewport, and add llms.txt. Most sites can move from Significant gaps to Solid foundation with two to four weeks of focused technical work.
Is the score guaranteed accurate?: No. The score is a systematic estimate based on observable signals, not a direct measurement of AI citation frequency. It does not guarantee that a high-scoring site will appear in any specific AI answer, or that a low-scoring site will not. AI search ranking is influenced by many factors PromptScore does not measure, including the AI provider's own content policies, query intent, and real-time retrieval decisions. The value of the score is its repeatability: the same inputs reliably produce the same score, so you can measure progress.

See how your site scores

Run a free scan against all 36 checks. Results in under two minutes.

Scan my site