The audit found the gaps. This closes them.

Before authority matters, AI engines have to crawl you, render you, index you, and be allowed to quote you. Five deterministic checks decide whether anything else in this playbook can work.

Why it carries this weight: It is a cap, not a pillar: a hard failure here nullifies every dollar of authority spend. Most modern SaaS sites fail silently on one item — usually JavaScript-only rendering that AI crawlers never execute.

• →Open your robots.txt and confirm GPTBot, PerplexityBot, Google-Extended, BingBot, and CCBot are not blocked. This is a five-minute check that decides citability outright.
• →Fetch your key pages with JavaScript off. If the body text isn’t in the raw HTML, prerender or server-render those routes. Marketing pages first; the app can stay client-side.
• →Verify indexation in both Google and Bing. ChatGPT search leans on Bing’s index, and almost nobody submits there. Search Console and Bing Webmaster, sitemap submitted in both.
• →Hunt down stray noindex and nosnippet directives on pages you want quoted. A nosnippet tag is a standing instruction to AI Overviews to ignore you.
• →Add an llms.txt and keep your sitemap honest, so engines that look for guidance find it.

Original research, benchmarks, and indexes that third parties cite. The single heaviest investment in the model, because being the source beats being a citer everywhere downstream.

Why it carries this weight: AI engines synthesize from sources they trust, and nothing earns source status like data nobody else has. One benchmark report generates citations for years; ten blog posts generate none.

• →Pick one number your category argues about and own it: run the survey, publish the benchmark, name it like a product ("The [Category] Benchmark 2026").
• →Publish the methodology openly. Checkable methods are what separate citable research from content marketing wearing a lab coat.
• →Cut the data into ten assets: the report, a summary page per finding, charts people can lift (with attribution baked in), and a press-ready stats sheet.
• →Refresh annually on a named cadence. Year two is when journalists and LLMs start treating it as the standing reference.
• →Seed it to the people who cite things: analysts, newsletter writers, community moderators. Ask for nothing except a look.

Whether AI knows exactly who you are: one consistent identity across your site, schema, third-party databases, and the knowledge graph, cleanly separated from anyone with a similar name.

Why it carries this weight: Engines decide what you are before they decide whether to recommend you. An ambiguous entity gets merged, misattributed, or skipped, and every other pillar leaks through that hole.

• →Write the one-sentence canonical description and use it verbatim everywhere: site, LinkedIn, directories, press boilerplate. Sameness is the signal.
• →Ship Organization and Person JSON-LD with sameAs links connecting your profiles into one graph.
• →Claim the corroborating records: Crunchbase, G2, key directories. Wikipedia only if you genuinely clear notability — never manufacture it.
• →If you share a name with anyone (a newsletter, an agency, a band), add a disambiguating description and make every mention "Brand — what it is, by whom."
• →Map the three to five topics you want attached to your entity and reinforce them in every bio, byline, and about page.

A named human with a point of view: founder or executive voice on podcasts, stages, and feeds, attached to the topics the company wants to own.

Why it carries this weight: AI trusts people before companies. A strong personal entity lends its credibility to the brand entity, and quotable humans get cited where faceless brands get summarized.

• →Pick one POV worth disagreeing with and have the exec say it everywhere, in the same words. Repetition is how a take becomes an association.
• →Podcast tour over guest posts: transcripts feed the models, and hosts do the distribution.
• →Publish under a real byline with a real bio and Person schema. Anonymous "team" posts build nothing.
• →One platform, owned: pick where the audience lives and show up weekly, not everywhere monthly.
• →Turn every appearance into assets: clips, quotes, a transcript page on your domain.

Earned mentions in places engines already trust: trade press, analyst notes, industry newsletters, and the communities where your buyers compare notes.

Why it carries this weight: Citations are how engines triangulate truth. Mentions don’t need links to count anymore — the brand appearing in trusted contexts, consistently described, is the signal.

• →Run POV-driven PR, not feature announcements. Strong opinions create citations; release notes create silence.
• →Make your research (chapter 01) the pitch: data gives journalists a reason that "we shipped a feature" never will.
• →Build the analyst relationships in your niche — the mid-size voices who actually publish beat the big firms who gate everything.
• →Treat newsletters and curators as tier-one press. Their archives are exactly what models read.
• →Respond fast to journalist requests in your lane; reliability compounds into being the default quote.

Expert-led video whose transcripts become AI source material: teaching content, not promos, published where engines and buyers both look.

Why it carries this weight: Transcripts are text the models read, and video is the format buyers trust. One strong video becomes ten assets — the fan-out engine — and every derivative reinforces the same entity and topics.

• →Lead with expertise: teach the thing you want to be known for; the product appears only as evidence.
• →Publish transcripts on your own domain, structured with real headings — the video feeds YouTube, the transcript feeds everything else.
• →Run the 1→10 fan-out on every flagship video: clips, posts, quotes, a written companion piece.
• →Name videos like questions buyers actually ask; YouTube is the second-largest search engine and a primary LLM source.
• →Consistency over production value: a weekly teaching video beats a quarterly cinematic one.

Owning the exact questions buyers ask AI: definitions, comparisons, methodologies, and the "best X for Y" pages — written to be retrieved, not just read.

Why it carries this weight: Engines answer questions by retrieving passages. If your pages aren’t structured as answers — chunked, entity-explicit, question-first — someone else’s are.

• →List the 25 questions your buyers actually ask an AI (sales calls and community threads are the source). One page each, answer-first.
• →Structure for retrieval: H2/H3 chunks that stand alone, the entity named in every section (no orphaned "it"), scope made explicit.
• →Write the honest comparison pages, including against competitors. If you don’t define the comparison, a third-party affiliate will.
• →Add FAQ and HowTo schema where it genuinely fits the content.
• →Cover the counterarguments. Engines summarize both sides; one-sided pages lose the citation to whoever covers the objection.

Standing measurement of how engines treat you: recommendation frequency, citation frequency, competitive position, and narrative accuracy, tracked on a monthly cadence.

Why it carries this weight: You can’t improve what you can’t see, and AI answers drift without telling you. Five percent of effort, but it’s the pillar that makes every other pillar accountable.

• →Fix a panel of 10–15 blind category questions and run them monthly across ChatGPT, Claude, Gemini, and Perplexity. Same questions, every month — the trend is the data.
• →Log mentions, position, sentiment, and who else gets named. The competitor field tells you who’s winning attention you’re not.
• →Watch narrative accuracy: when engines describe you wrongly, trace it to the source and fix the source.
• →Re-run the full audit quarterly; tie pillar movements to the work shipped that quarter.
• →Purpose-built listening tools (Profound, Scrunch, Athena) when spreadsheet tracking outgrows itself.