Personal SEO Audit

A developer released a self-contained prompt for auditing a person's complete online presence, covering technical SEO, structured data, and content gaps. The tool includes optional integration with the open-source claude-seo skill and emphasizes evidence-backed diagnostics over assumptions.

You are auditing the complete online presence of one person: NAME . Not just their website — everything a search engine, an AI assistant, or a stranger doing due diligence would find. Your deliverable is an honest, evidence-backed report with a prioritized action plan. You change nothing; you only measure, diagnose, and recommend. This prompt is self-contained, but if you are a Claude Code-style agent with shell access, first install the open-source claude-seo skill MIT for deeper site tooling — it is optional, not required: git clone --depth 1 https://github.com/AgriciDaniel/claude-seo.git && bash claude-seo/install.sh Do 15 minutes of orientation searching so your questions are informed, not generic: - Web-search the person's name, name + employer, name + their field. - Identify: their personal site if any , their highest-ranking profiles, what the top-10 results collectively say they do , and roughly when their owned content was last updated. - Note any name collisions — other people with the same name who rank, and whether they're in an adjacent field adjacent collisions are the dangerous kind . Present these, pre-filled with your recon guesses so they can confirm/correct rather than write essays: Identity: What do you want to be known for in one sentence? Has that changed in the last 5 years? Old identities often dominate old records. Audience: Who do you most need to find you and come away impressed — employers, LPs/investors, customers, journalists, conference organizers, other? Surfaces: Confirm the full inventory — personal site + repo and platform: Jekyll/Hugo/Squarespace/etc. , X/Twitter, LinkedIn, GitHub, Medium/Substack, YouTube, podcast host or frequent guest? , books, employer bio page, speaker profiles, anything else. Which do you control login access to? Ground truth: Ask them to run an incognito Google search of their own name and paste a screenshot or description of the first page including any Knowledge Panel. Do not skip this. Your tool-mediated search results are NOT what Google actually serves; auditing from them alone will produce confidently wrong entity findings. Constraints: Anything off-limits old content they want left alone, platforms they refuse to use, facts they don't want published — e.g. AUM, salary, location ? Effort budget: Are they willing to do a weekend of fixes, or only delegate to an agent/assistant? Incorporate the answers, then run everything below. Crawl and measure — produce numbers , not impressions. For every claim, keep the evidence commands run, counts, URLs . Technical: robots.txt , sitemap.xml fetch it, parse it, count URLs, check lastmod dates — the newest lastmod tells you when the site actually last changed .- HTTP→HTTPS and www/apex redirects; canonical tags; response headers HSTS, X-Content-Type-Options, Referrer-Policy ; TTFB. - Crawl every sitemap URL programmatically. Per page record: status code, <title length , meta description presence/length , H1 count, image count, images missing alt, word count. Report aggregates: broken pages, pages with 0 H1s, titles 60 chars, missing descriptions, thin pages <400 words , median word count, % images without alt. - Internal links on the homepage and nav — test every one for 404s dead nav links are common on stale sites and are on every page . - RSS/Atom feed: present? Last entry date? Structured data: Extract all JSON-LD. Validate types against schema.org — watch for case errors Website vs WebSite , a Person modeled wrongly as publisher / Organization , missing Person entity, missing sameAs array, missing Article / BlogPosting with author + dates. For a personal brand, Person + sameAs is the single highest-leverage block on the site. Performance: Try PageSpeed Insights API; if quota-blocked, measure directly: total CSS/JS weight, render-blocking resources, legacy frameworks jQuery/Semantic/Bootstrap on a text blog , image formats and sizes, third-party embeds Disqus, analytics . Freshness: Date of newest post; date of last build last-modified headers ; any page that announces its own staleness "this list is no longer updated…" . These pages tell machines the person stopped. Measurement: Any analytics tag? Google Search Console verification meta tag? If neither, flag: they are flying blind and that's a Phase-1 fix. This is where personal-brand audits are won or lost, and where you must be most careful about evidence quality. Knowledge Panel: From the human's incognito screenshot Phase 0.5 : does a panel exist? What is the subtitle/entity type ? "Internet personality" or a stale occupation = high-priority finding. What does the Overview paragraph say and — critically — what source is it attributed to ? Google Books bios from old books are a classic stale-Overview source. Note: the Amazon profile card is fed by Amazon Author Central, but a Google Books Overview comes from the book's publisher metadata — two different records, two different fixes. Which profiles does the panel link, and which major ones are missing ? Wikidata: Query the API wbsearchentities for their name. Does an item exist? Is it them or a name-twin check before recommending edits ? Wikidata is the documented lever for a panel's occupation subtitle, and creating an item about yourself is permitted unlike Wikipedia autobiography . Wikipedia: Article? If not, assess notability honestly — and always recommend against self-authored or paid articles deletion debates are worse search results than absence . Third-party knowledge sources: Whatever LLMs and journalists actually ground on for this person's field — Crunchbase, Bloomberg profiles, IQ.wiki, IMDB, Google Scholar, industry wikis. Check each for staleness and factual errors; these are correctable and heavily read by machines. Name collisions: Who else shares the name? Do they rank on page one per the human's SERP, not yours ? Do they hold the "good" handles linkedin.com/in/firstlast, etc. ? Books/products: Amazon Author Central claimed? Goodreads? Old bios on those records leak into the Knowledge Graph. For each platform, record: active/dormant, last activity date, follower/reach numbers, what the bio says, and whether it links back to the owned site. Cover at minimum: X/Twitter — including whether the display name is machine-matchable to their real name stylized names break entity resolution , the pinned post, and the Articles and Highlights tabs native long-form that may be portable to the owned site . LinkedIn — usually a top-2 result for any professional's name. Read the summary's first sentence : is it their current identity or a former one? GitHub, Medium, Substack, YouTube, podcasts host or recurring guest — a weekly show absent from the personal site is a huge unclaimed asset , speaker pages, employer team page. Employer/organization site: Does the person have a stable, server-rendered bio URL there not a JS accordion ? Does it link out to their properties? A firm site that passes no link equity to its own people is a free fix. The ownership test: Where do their best recent essays/ideas actually live? If the pattern is "published as threads/posts on rented platforms, then rewritten by aggregators who capture the search traffic and citations," that is typically the single largest value leak. List the specific pieces worth back-publishing to the owned domain with canonical URLs, original dates, and provenance notes. The lede test: Read sentence one of every bio across every surface. Do they all lead with the current identity, or does an old chapter the startup that sold, the sport, the previous career still occupy the lede everywhere? Old chapters are hooks — they belong in sentence two, not the headline machines read as identity. Crawler access: Fetch key pages with the actual user agents — GPTBot, ClaudeBot, PerplexityBot, Google-Extended — and record status codes. Check robots.txt for AI-crawler blocks. Check for llms.txt . Groundability: Access is necessary, not sufficient. When a model answers "who is NAME " or "what does NAME think about their topic ," it needs a source that is first-party, current, and structured. Assess: does such a page exist? If not, models fall back on third-party profiles — identify which ones ask 2–3 AI assistants about the person and note which sources and which era of the person they reflect . Citability: Are their signature ideas on canonical URLs a model can cite, or trapped in threads/videos/podcasts with no text home? Score 0–100, weighted : Technical 22%, Content quality 23%, On-page 20%, Schema 10%, Performance 10%, AI-search readiness 10%, Images 5%. Score from your measurements; say explicitly what you couldn't measure. Findings, ranked by consequence not by ease . For each: the evidence numbers/URLs , why it matters for this person's stated audience, and a falsifiability check — "here is how you'd know this diagnosis is wrong" or "here is how you'll know the fix worked, and by when." If you can't state one, the finding is an opinion; demote it. Action plan in phases: a weekend highest-leverage, self-serve , 30 days, 90 days, ongoing habits. For each item: a leading indicator the person can watch without re-auditing. Sequence by dependency e.g., Search Console before content changes; schema before expecting panel changes; source-record fixes before expecting Overview changes, with realistic 2–12-week propagation windows . A "what NOT to do" section. Typical entries: don't write your own Wikipedia page; don't chase generic head terms in your industry you'll lose to publishers and don't need them — your queries are your name and your ideas ; don't rebuild the site a rewrite is procrastination dressed as work ; don't add deprecated schema FAQ rich results are dead; HowTo is dead ; don't fix the ownership problem by posting more on rented platforms. Never present tool-mediated search results as the real SERP. Rankings you see through an API differ from what Google serves a human. Where it matters — especially entity findings — rely on the human's incognito screenshot, and mark all other ranking claims as indicative. Never invent content or facts. Figures that vary across sources AUM, follower counts, dates get a CONFIRM placeholder and a list at the end. If you can't recover a text, say so; never reconstruct from memory. Report what you couldn't measure API quotas, blocked fetches and how that limits confidence. If the human's ground truth contradicts your finding, the finding changes — and say so visibly in the report. A corrected audit is more credible, not less.- Every recommendation carries its falsifiability check. No unfalsifiable advice. A single report Markdown or HTML with: a one-paragraph verdict up top, the scorecard, ranked findings with evidence and falsifiability checks, the surface-inventory table, the phased action plan, the what-not-to-do list, and a methods/limitations footer. If asked, follow with a second document: implementation instructions an agent with site access could execute — all changes staged in a PR for review, never pushed live, and all third-party-platform updates drafted as copy for the human to paste, never executed by the agent.