The audit library

Every page NarrativeBot crawls is judged against this library. A check is either per page, reading that page’s own response, or whole site, reading the link graph and the site files, because a page cannot know what links to it, how deep it sits, or that another page is a copy of it. Each check produces a finding that says what it found, what it expected, and its severity. Findings on the same family and page roll up into one piece of work rather than a dozen tiny ones, and a family is the unit a Repair is opened on.

Status says whether the check runs today. A check marked Coming is in the catalogue and records nothing yet, rather than a fake pass. The library grows by release; this page is kept in step with the code.

Reading the table

Column	Meaning
Check	The display name and the key you will see in the API and in exports.
What passes	The rule the page or site is held to.
Severity	Gate, High, Medium or Low. See severity. Gates are the two faults that make a brand invisible to an engine outright.
Scope	Per page, or whole site (needs the full crawl and the link graph).
Engines	Who the check matters for. Answer engines means the non-Google crawlers (OAI-SearchBot, PerplexityBot, Claude-User and their peers), which do not run JavaScript.

Technical checks

The technical half of the library covers whether an engine can fetch, render, read and trust the page. Two of these are gates.

Crawler access

Check	What passes	Severity	Scope	Engines	Status
AI crawler access in robots.txt `ai_bot_access`	robots.txt must allow OAI-SearchBot, PerplexityBot, ClaudeBot, and Google-Extended. Blocking any of these makes the brand invisible to that engine.	Gate	Whole site	Search and answer engines	Live
CSS/JS resources unblocked for Googlebot `resources_unblocked`	robots.txt does not block CSS or JS files Googlebot needs to render the page correctly.	Medium	Whole site	Google	Live

Renderability

Check	What passes	Severity	Scope	Engines	Status
SSR without JavaScript `ssr_no_js`	Critical page content must be present in the raw HTML response with JavaScript disabled. Non-Google engines (ChatGPT, Perplexity, Claude) do not execute JS.	Gate	Per page	Answer engines	Live
Answer is extractable from plain DOM text `extractable`	Answer content lives in plain DOM text: not an image, an accordion-hidden section, or behind a login wall. Extractors cannot read what they cannot see.	High	Per page	Search and answer engines	Live

Indexability

Check	What passes	Severity	Scope	Engines	Status
Indexable with snippet allowed `indexable_snippet`	Page is indexed and does not carry noindex, nosnippet, or max-snippet:0. Without snippet permission AI Overviews cannot quote the page.	High	Per page	Google	Live
XML sitemap submitted to GSC `xml_sitemap`	An XML sitemap listing canonical URLs is submitted and healthy in Google Search Console.	Medium	Whole site	Google	Live
Correct HTTP status codes `http_status`	Pages return correct status codes: 200 for indexable, 301/302 for redirects, 401 for auth, 404 for missing. Only 200 pages are indexed.	Medium	Per page	Google	Live
SPA not-found emits real 404 or noindex `no_soft_404_spa`	Single-page-app ‘not found’ states emit a real HTTP 404 or a noindex meta tag, not a 200 with an error message (soft-404).	Medium	Per page	Google	Live
Google Search Console verified `gsc_verified`	GSC property is verified via HTML tag, DNS, GA, or GTM. Prefer DNS domain-property for full-site coverage. Token is persistent and reachable without auth.	Medium	Per page	Google	Coming

Canonical URLs

Check	What passes	Severity	Scope	Engines	Status
Self-referencing canonical link `canonical`	Each indexable page carries a self-referencing rel=canonical and duplicate-content variants are minimized.	Medium	Per page	Search and answer engines	Live
JS-injected canonical matches original `canonical_js_consistent`	Any canonical URL injected by client-side JS matches the server-rendered canonical. Only one canonical per page.	Medium	Per page	Google	Coming

Metadata

Check	What passes	Severity	Scope	Engines	Status
Unique, descriptive title tag `title_tags`	Every indexable page has a unique, descriptive `<title>`. Length in the 30-60 char range so it renders without truncation in Google’s SERP.	Medium	Per page	Google	Live
Unique meta description `meta_description`	Every indexable page has a unique 1-2 sentence meta description. Not a ranking factor, but drives snippet quality when Google uses it.	Medium	Per page	Google	Live
Titles are unique `duplicate_titles`	No two pages should share a title. A repeated title splits the signal between the pages that carry it and gives a reader no way to tell them apart in a list of results.	Medium	Whole site	Search and answer engines	Live
Meta descriptions are unique `duplicate_meta_descriptions`	No two pages should share a meta description. A repeated description describes at most one of the pages using it accurately.	Medium	Whole site	Search and answer engines	Live

URL structure

Check	What passes	Severity	Scope	Engines	Status
Word-based URLs with topical directories `descriptive_urls`	URLs use human-readable words and organize content into topical directories rather than opaque IDs.	Medium	Per page	Google	Live
History-API routes, not fragment URLs `history_api_routing`	Routes use real /path URLs via the History API, not #/fragment paths that Googlebot ignores in indexing.	Medium	Per page	Google	Live

Link hygiene

Check	What passes	Severity	Scope	Engines	Status
No orphan pages `orphan_pages`	Every page must be reachable by at least one internal link. A page nothing links to is found only by someone who already has the URL, receives no authority from the rest of the site, and is invisible to any crawler that discovers by following links.	High	Whole site	Search and answer engines	Live
Internal links resolve `broken_internal_links`	Internal links must point to pages that load. A link into a missing page is a dead end for a reader and a wasted fetch for a crawler. Reported against the page holding the link, which is the part that can be edited.	High	Whole site	Search and answer engines	Live
Crawlable `<a href>` navigation `crawlable_links`	Site navigation and links use `<a href>` to real URLs with descriptive anchor text: not JS-only click handlers.	Medium	Per page	Google	Live
No redirect chains `redirect_chains`	An internal link should reach its destination in at most one redirect. Longer chains lose signal at every hop and are abandoned by crawlers before the end; a loop never resolves.	Medium	Whole site	Search and answer engines	Live
Pages are not buried `click_depth`	A page should sit within a few clicks of the homepage. The deeper a page is, the less often it is crawled and the less internal authority reaches it.	Medium	Whole site	Search and answer engines	Live
nofollow/ugc/sponsored on untrusted links `ugc_nofollow`	User-generated and paid outbound links carry rel=nofollow, rel=ugc, or rel=sponsored as appropriate.	Low	Per page	Google	Live
Internal authority reaches every page `internal_authority`	A page should receive a meaningful share of the site’s internal link authority. A page linked only from places that carry no weight is technically reachable and practically buried.	Low	Whole site	Search and answer engines	Live

Entity identity

Check	What passes	Severity	Scope	Engines	Status
Organization schema with sameAs `entity_sameas`	JSON-LD Organization block with sameAs links pointing to Wikidata, Wikipedia, and LinkedIn. Anchors the brand as a resolvable entity across engines.	High	Per page	Search and answer engines	Live
Brand, product, category named consistently `entity_clarity`	Brand, product, and category names are used explicitly and consistently across the page. Enables extractor attribution to the correct entity.	High	Per page	Search and answer engines	Coming

Image semantics

Check	What passes	Severity	Scope	Engines	Status
Descriptive alt text on images `image_alt`	Images carry descriptive alt text, sit near relevant text context, and use meaningful filenames.	Medium	Per page	Google	Live

Freshness

Check	What passes	Severity	Scope	Engines	Status
Freshness signals aligned with real updates `freshness`	Visible ‘last updated’ date matches the JSON-LD dateModified value and reflects a real content delta. Non-Google engines (Perplexity especially) prioritize fresh.	Medium	Per page	Answer engines	Live

Performance

Check	What passes	Severity	Scope	Engines	Status
Good Core Web Vitals / fast TTFB `core_web_vitals`	Lighthouse Core Web Vitals in the ‘Good’ band and TTFB under Google’s threshold. Slow pages get deprioritized in Google’s AI answer selection.	Medium	Whole site	Google	Live

Rendering reliability

Check	What passes	Severity	Scope	Engines	Status
Lazy-loaded content renders without scroll `lazy_load_safe`	Lazy-loaded content becomes visible without user scroll (intersection observer + prerender-friendly). The renderer does not scroll.	Medium	Per page	Google	Live
Fingerprinted JS/CSS asset filenames `asset_fingerprinting`	JS and CSS files use content-hash filenames so the renderer cache does not serve stale assets after a deploy.	Low	Per page	Google	Live

Policy compliance

Check	What passes	Severity	Scope	Engines	Status
Passes Google spam policies `spam_policy`	Site passes Google’s spam policies: no scraped or scaled content, no cloaking, no keyword stuffing, no auto-generated gibberish.	High	Per page	Google	Coming
No intrusive interstitials `no_interstitials`	Main content is not blocked by intrusive interstitials (popups covering the fold, forced app-install prompts).	Medium	Per page	Google	Live

Discovery

Check	What passes	Severity	Scope	Engines	Status
IndexNow protocol enabled `indexnow`	Site publishes IndexNow keys. Bing / Copilot pick up content changes faster with IndexNow than via crawl only.	Low	Per page	Bing and Copilot	Coming

Content checks

The creative half judges whether a page is written to be quoted: an answer up front, concrete sourced figures, consistent naming, and passages that stand on their own. These use a model to read the page, so they are being switched on family by family as each is validated. Where the audit is not yet running a check it is marked Coming and records nothing.

Content quality

Check	What passes	Severity	Scope	Engines	Status
No near-duplicate pages `near_duplicate_content`	No page should be substantially a copy of another. Pages whose bodies differ only by a name or a city compete to answer the same question, and an engine picks one and discards the rest.	High	Whole site	Search and answer engines	Live
Free of spelling and grammar errors `spelling_grammar`	Content is free of spelling and grammar errors. Google’s ‘compelling and useful’ guidance treats this as a baseline quality signal.	Medium	Per page	Search and answer engines	Coming

Structure and extractability

Check	What passes	Severity	Scope	Engines	Status
Direct answer in the first ~50 words `answer_first`	Page opens with the core answer in the first ~50 words. ~44% of AI citations come from the first ~30% of the page.	High	Per page	Search and answer engines	Live
Real questions as headings `qa_format`	Real user questions used as section headings with concise answers immediately below. Maps directly to conversational AI queries.	Medium	Per page	Search and answer engines	Coming
Self-contained passages `standalone_passages`	Sections are self-contained enough to stand alone when extracted. Improves non-Google extractability. NOT a Google ranking requirement.	Medium	Per page	Answer engines	Coming

Evidence and citations

Check	What passes	Severity	Scope	Engines	Status
Concrete, sourced statistics `quotable_stats`	Content includes concrete statistics with named sources. Princeton GEO research finds this is the largest measured lever (+41% citation rate).	High	Per page	Search and answer engines	Coming
Primary sources + named-expert quotes `cited_sources`	Content cites primary or authoritative sources and quotes named experts. Boosts citation rate ~28-40%.	High	Per page	Search and answer engines	Coming
Outbound links to trusted sources `corroborating_links`	Content links outward to trusted, authoritative sources (untrusted sources should carry nofollow). Demonstrates topical knowledge to Google.	Medium	Per page	Search and answer engines	Coming

Topic coverage

Check	What passes	Severity	Scope	Engines	Status
Comparison / ‘X vs Y’ / decision tables `comparison_content`	Topic-relevant comparison content: ‘best X for Y’, ‘X vs Y’, or decision tables. Highest-citing content format for AI answer surfaces.	High	Per page	Search and answer engines	Coming
Coverage of competitor-owned angles `topic_coverage_gap`	Page/site covers topics and angles that competitors already own for the target Topic. NB: means depth on Topics the brand should own: not spammy variation-coverage which trips Google’s scaled-content-abuse policy.	High	Per page	Search and answer engines	Coming

Originality

Check	What passes	Severity	Scope	Engines	Status
Original data, testing, or POV `first_hand_value`	Page contains original data, testing, or an opinionated point of view: not a commodity restatement of what’s already common. Google says this influences AI visibility more than any other factor.	High	Per page	Search and answer engines	Coming

Supporting media

Check	What passes	Severity	Scope	Engines	Status
Relevant images / video with SEO metadata `supporting_media`	Pages include relevant images or video that follow image/video SEO best practices. AI surfaces media as additional visibility slots.	Medium	Per page	Search and answer engines	Coming

External signals

Check	What passes	Severity	Scope	Engines	Status
Google Business Profile / Merchant Center `gbp_merchant`	Where relevant to the brand type: an active Google Business Profile and Merchant Center feed. Both feed local/product info into AI answers.	Low	Per page	Google	Coming

Who the crawler-access check looks for

The AI crawler access gate reads your robots.txt for the agents that matter, grouped by what blocking each one costs:

Group	Agents	Cost of blocking
Retrieval	`OAI-SearchBot`, `PerplexityBot`, `Claude-User`	These fetch a page because somebody asked a question right now. Blocking one removes you from that engine’s answers immediately.
Training	`GPTBot`, `ClaudeBot`, `CCBot`, `Google-Extended`	These feed future models. Blocking costs nothing today and compounds quietly.
Search	`Googlebot`, `Bingbot`	The classic pair. Most AI answers rest on the search index underneath them.

The finding names which group is blocked, because “invisible in ChatGPT today” and “missing from the next model” call for different urgency.

How findings become work

A finding on a page is one row in Site Health, under the page and under the issue.
Findings in the same family on a page are one opportunity, not several.
Opening a Repair on a family creates one Fix per affected page. The next crawl re-judges each Fix; one the audit no longer sees closes itself.

Getting started

Connectors

Technical SEO

Release notes

Reading the table

Technical checks

Crawler access

Renderability

Indexability

Canonical URLs

Metadata

URL structure

Link hygiene

Entity identity

Image semantics

Freshness

Performance

Rendering reliability

Policy compliance

Discovery

Content checks

Content quality

Structure and extractability

Evidence and citations

Topic coverage

Originality

Supporting media

External signals

Who the crawler-access check looks for

How findings become work

​Reading the table

​Technical checks

​Crawler access

​Renderability

​Indexability

​Canonical URLs

​Metadata

​URL structure

​Link hygiene

​Entity identity

​Image semantics

​Freshness

​Performance

​Rendering reliability

​Policy compliance

​Discovery

​Content checks

​Content quality

​Structure and extractability

​Evidence and citations

​Topic coverage

​Originality

​Supporting media

​External signals

​Who the crawler-access check looks for

​How findings become work

Reading the table

Technical checks

Crawler access

Renderability

Indexability

Canonical URLs

Metadata

URL structure

Link hygiene

Entity identity

Image semantics

Freshness

Performance

Rendering reliability

Policy compliance

Discovery

Content checks

Content quality

Structure and extractability

Evidence and citations

Topic coverage

Originality

Supporting media

External signals

Who the crawler-access check looks for

How findings become work