I Audited My Own SEO Site for AI Visibility — Here’s What I Found (And What’s Broken)

Most “AI SEO” content right now is theory. Agencies are selling six-figure GEO retainers without ever publishing a single end-to-end audit of a real site — including their own. So I ran the full AI visibility audit on seorevive.in — the site you’re reading this on — and I’m publishing what I found, broken parts and all.

The honest preamble

A few things to get out of the way before the findings:

• I built seorevive.in with AI citation in mind from day one. Schema, llms.txt, robots.txt allowlist for AI crawlers, content structure — all designed for it. So this isn’t a “before” picture of a typical site. It’s a stress test of what a technically well-prepared site still needs.
• I used Claude (Opus 4.7) for the heavy lifting — schema extraction, content quotability scoring, JSON-LD generation. I did the synthesis, the judgement calls, and this writeup. That mirrors how I’d run a paid audit for a client: AI does the production, I do the expertise.
• I did not run live LLM citation testing in this audit. That’s Phase 2. This is the on-site readiness audit — the part that can be done from a URL alone.

The 4-phase framework I use

Every AI visibility audit I run covers four phases. The first three are on-site and can be done from the URL alone. The fourth requires browsers or APIs.

1. Schema audit — what structured data exists, what’s missing, what’s broken
2. Content quotability audit — which pages have liftable content that LLMs prefer
3. Author authority audit — does the brand have a coherent entity graph
4. Live LLM citation testing — query the AI engines, see who’s cited (Phase 2)

For each phase, I score on a 0–10 scale. Here’s where seorevive.in landed:

Overall AI visibility readiness: 6.5 / 10.

That’s a real, measured score. Not a sales pitch. The headline finding is uncomfortable: I built a technically excellent site, but it has almost no signal to AI engines that it exists as an entity worth citing. Let me explain.

Finding 1: The schema is strong (and most sites get this wrong)

When LLMs decide what to quote, they prefer extracting from pages with rich structured data. Not just “have a schema” — but the right schema for the content type.

What I had on seorevive.in:

• Organization with knowsAbout (17 sub-topics), hasOfferCatalog (5 services with prices), sameAs
• ProfessionalService with geo coordinates, areas served, service types
• Person for me with knowsAbout, knowsLanguage, workLocation, nationality
• Article, FAQPage, BreadcrumbList, Service schemas distributed across pages
• Explicit AI bot allowlist in robots.txt (GPTBot, ChatGPT-User, OAI-SearchBot, etc.)
• Curated /llms.txt per the proposed llmstxt.org spec

What was missing:

• HowTo schema on the HCU recovery playbook. The post is a how-to (90-day, phased recovery framework). Without HowTo schema, it’s just an article — AI engines can’t extract the steps cleanly.
• WebSite schema with SearchAction. Tells AI engines how the site is structured and searchable.
• SpeakableSpecification on FAQ answers. Boosts voice and AI-summary readiness.
• ImageObject metadata on embedded charts. Without it, your custom diagrams are invisible to AI extractors.

These are 10–30 minute fixes per item. I’m shipping them this week.

The lesson for any site: rich schema is necessary but not sufficient. And the missing schema types matter more than you’d expect — HowTo specifically is a citation magnet for “how do I X” queries, which is a huge slice of what people ask LLMs.

Finding 2: The flagship content is citation-ready

Of the 7 blog posts on the site, the HCU recovery playbook scores highest on quotability — 8/10. Here’s why it works for AI citation:

1. It cites specific dates with sources. March 2024 (when HCU was folded into the core algorithm). May 2024 (the API leak). August 15 2024, December 12 2024 (specific core updates). LLMs love dates because they reduce hallucination risk when re-quoting.
2. It names specific Google attributes from the leak. siteAuthority, siteFocusScore, hostAge, contentEffort — these are unique enough strings that an LLM citing them needs a source.
3. The “In plain English” callout box is structured for lifting. Four definitive bullets, each a self-contained statement. That’s exactly the pattern AI extractors prefer.
4. The FAQ has six fully self-contained Q&As. Each answer stands alone. An LLM can quote one without needing surrounding context.
5. The lead has a contrarian thesis. “Most recovery advice you’ll find online was written before that change” is a strong opinion claim. LLMs surface contrarian positions more than safe ones because they’re more informative.

The other six posts? They’re decent traditional SEO content. But for AI citation specifically, they need work — fewer definitive statements, fewer stat anchors, no “in plain English” structures, weaker FAQs. That’s the next 60 days of content upgrades for me.

If you’re auditing your own content for AI citation, here’s the test: pick one paragraph at random. Could an LLM quote it word-for-word and have it stand alone? If yes, that paragraph is citable. If no — if it depends on surrounding context or hedges every claim — rewrite it.

Finding 3: The author entity is the rate-limiting step

Here’s where it got uncomfortable.

My Person schema is rich. knowsAbout has 13 SEO sub-topics. knowsLanguage covers English, Hindi, Assamese. workLocation spans India, US, Canada. I look like a credible expert on paper.

But my sameAs array has exactly one entry: LinkedIn.

That’s the entity authority gap. AI engines build their understanding of “is this person a real expert” by following the sameAs graph — the cross-references that prove identity continuity across the web. With one entry, I look like a brand-new persona with no track record, regardless of whether I have 8 years of real experience.

What I should have (and don’t):

• Twitter/X profile (huge miss — Twitter is heavily cited by Perplexity)
• Medium profile (for syndication)
• Substack or newsletter (none yet)
• YouTube channel (planned, not started)
• Crunchbase entry (free to create, takes 20 minutes)
• Bylines on Search Engine Journal, Search Engine Land, MOZ, HubSpot, Semrush blog
• Wikidata entry (Q-number — lower notability bar than Wikipedia)

This is the single biggest deficit on the site. Without external profiles confirming my identity, the rich Person schema is hanging in the air with nothing to anchor to.

If you have a personal brand inside your SEO or marketing site, audit your sameAs graph right now. Count the entries. If it’s under 5, you’re invisible to AI engines as a real person.

Finding 4: Off-site is the real bottleneck

This was the hardest finding to write honestly.

I scored 2/10 on off-site entity authority. That’s because — despite a brand-new, technically excellent site — seorevive.in has:

• Zero verified directory listings (Clutch, GoodFirms, Sortlist not yet submitted)
• Zero brand mentions in Reddit or Quora threads that LLMs already cite
• Zero reviews on G2, Capterra, Trustpilot, or anywhere else
• Zero press mentions (HARO is dead; I haven’t pursued the live alternatives yet)
• Zero podcast appearances
• Zero published bylines outside my own site

When ChatGPT decides what to cite for “best SEO agency in India for small business,” it pulls from sources it knows. My site, despite being technically sophisticated, doesn’t yet appear in those sources. The entity isn’t recognised yet.

This is the unsexy truth about AI visibility: 70% of the work is on-site, but 100% of the citation happens off-site. You can have perfect schema and never be cited if your brand isn’t distributed across the surfaces LLMs trust.

This isn’t unique to my site. It’s the default state of every new domain. The fix is the slow work: directory submissions, real participation on Reddit and Quora, guest posts on top-tier publications, podcast appearances. Months of grinding. There’s no schema-only shortcut.

What I’m shipping in the next 30 days

I made a prioritised list of 12 fixes. Here’s the queue, with effort estimates:

The first six are Tier 1 — all under 3 hours of total work and shipping this week.

What this means for your site

If you want to run this audit on your own site, here’s the framework:

Phase 1: Schema audit (45 minutes)

Open your homepage’s HTML source. Search for application/ld+json. List every schema type present. Check the schema.org documentation for what’s missing given your site type. Score 0–10.

Phase 2: Content quotability (1–2 hours)

Pick your top 5 pages by traffic. For each, count definitive statements, stat anchors, FAQ blocks, comparison tables. If they’re sparse, your content is decorative — pretty but not liftable.

Phase 3: Author authority (30 minutes)

If your site has named authors, find their Person schema. Count the sameAs entries. Under 5? You’re invisible as a person. Then check whether they have bylines on any cited publication.

Phase 4: Off-site reality check (15 minutes)

Open ChatGPT or Perplexity. Type “best [your category] in [your geo].” Are you mentioned? Type “[your brand] review.” What comes back? If you’re not in the results, your entity isn’t yet recognised.

That’s the whole methodology. Took me about 4 hours total on seorevive.in, including writeup.

What I’m not promising

A few honest disclaimers, since this is meant to be a practitioner audit and not a sales pitch:

• Schema fixes alone won’t make you appear in ChatGPT. They’re necessary but not sufficient. The off-site work is the real lever.
• AI citation takes 60–180 days minimum to reflect changes, because AI engines update their training and indices on slow cycles. Don’t expect overnight changes.
• “AI SEO” as a separate category will probably fade by 2027–28, the same way Mobile SEO and Voice Search SEO faded. Build durable SEO capability, not a brand around one buzzword.
• There’s no AI rank tracker I trust yet. The tools that promise daily citation tracking (Profound, Otterly, etc.) are too noisy for May 2026. Manual monthly testing in incognito tabs is more accurate.

What’s next

I’ll publish the follow-up — “60 Days After: Did the AI Citation Fixes Move Anything?” — once the changes have been live for two months. Baseline citation testing happens this week, before any fixes ship. That way the delta is measurable.

If you want a similar audit on your own site, you can reach out. I’m running a small number of free audits in May and June in exchange for permission to publish anonymised findings as a case study.

Frequently asked questions

About Nabin Thakur

Founder of SEORevive, an SEO consulting practice based in Guwahati, India. 8+ years of SEO experience, 100+ delivered campaigns across India, USA, and Canada. Specialises in Helpful Content Update recovery, technical SEO, e-commerce SEO, and AI search optimisation (AEO/GEO).

LinkedIn → · More about Nabin → · Work with him →