I Asked Claude to SEO-Audit My Own Site Live. It Found 10 Problems, and One of Them Was Embarrassing.

Let's get it out of the way.

Founder Signal is a directory of VCs, lawyers, accountants, and other service providers that founders can search. It's programmatic SEO by design: over a thousand pages, auto-generated from a database, each one a potential landing page for "best startup lawyer in Sydney" style queries.

Here's what I missed: most of those pages are thin. Some have reviews. Some have a company description. Many have neither, because the database is still filling in. Google's algorithm does not love a site that ships 1,000+ near-empty pages and asks to be ranked.

I'd indexed the lot. All of them. Live. For six weeks.

If you're running programmatic SEO on an AI-first site, this is the mistake to avoid: index on a quality threshold, not at launch. Auto-index pages once they pass a content bar (word count, review count, structured data populated), and no-index the rest until they grow up. Re-run the check weekly.

I'm setting the threshold this week. The shame is free.

Before you fix anything, get the honest picture. Here's the prompt I pasted live:

"Tell me about the SEO setup of my site. Describe it. Tell me what's good, what's bad, and what the improvements are. Give me lots of options but also your strict recommendations. Is it actually well set up? Is it ranking? What keywords should I be considering? What blind spots do I have? I'm not an SEO expert, guide me through."

Two things matter about that prompt:

1. "Give me strict recommendations." Without this, Claude returns a balanced "here are five options" menu that you have to synthesise. Ask for the verdict and it gives you one.
2. "I'm not an SEO expert." Grounds the output. Claude picks a level of jargon that fits, and surfaces things you'd miss if you pretended to know more.

You'll get a long report back. Screenshot it. Run through the flags one by one.

Here's the checklist in the order Claude surfaced them, cleaned up and decoded:

Some of these I'd actually fixed previously (the JSON-LD schema was already in place when I asked, I just couldn't remember). Most were real gaps. A few I'm still working through.

Two of them are doing the heavy lifting for why this is a different conversation than "classic SEO advice." Let me unpack those.

For twenty years the SEO world was "get more backlinks from trusted domains." The new version is entity authority.

Google's knowledge graph, Claude's training, ChatGPT's retrieval, Perplexity's index: they all want to confirm *who you are* before recommending you. If your company exists on:

• Wikipedia (the hardest and most valuable)
• Crunchbase (standard table stakes for anyone VC-adjacent)
• Google Business (local SEO + the map pack)
• LinkedIn Company Page (most already have this)
• Industry-specific databases (e.g. Clutch for agencies, G2 for SaaS)

...then the signals stack up and the models treat you as a real entity. Miss these and you look like a ghost.

For a six-week-old site, Wikipedia is probably unrealistic (the notability bar is high). Crunchbase, Google Business, and niche directories are easy wins. My plan: research the shortlist this week, apply for everything realistic, queue Wikipedia for a future where I have enough press coverage to clear notability.

If you're building an AI-first product right now, treat entity authority as week-one infrastructure. You'll need it for the next wave of search, and it compounds.

Here's the bit most founders haven't caught up with.

Every week, millions of people now ask ChatGPT, Claude, and Perplexity things like "what's the best tool for X" or "who runs the top newsletter on Y." Those answers cite specific sites. If your site gets cited, you get traffic (and increasingly, direct conversions). If you don't get cited, you're invisible to an entire channel that's growing faster than Google.

This channel has no SERP you can screenshot. You can't see your rank. Most founders don't even know they're missing it.

I've started running a weekly script that queries Gemini and Groq with 20 prompts about topics where I should plausibly be cited (startup advice, founder tools, pitch deck feedback). It logs whether batko.ai shows up in the responses and in what context. It's rough. It's also the only way I know to track this.

If you're spending an hour a week on Google rank tracking and zero hours on LLM visibility, you're optimising for a channel that's shrinking and ignoring a channel that's growing.

Concrete things that move LLM visibility:

1. Long-form, specific content (models retrieve quotable paragraphs)
2. Entity authority (see section above)
3. Direct answers to specific questions (the exact questions real founders type into ChatGPT)
4. Structured data (helps retrieval, helps citation confidence)
5. Topical consistency (the more focused your site, the easier to cite)

Most of these overlap with classic SEO best practice. The key shift is that specificity beats volume. A 3,000-word guide that answers one question thoroughly will outperform 20 thin pages every time.

Tempting to say: "Stop optimising for Google. Optimise for Claude, ChatGPT, and Perplexity."

That's not quite right. Google still drives the majority of traffic in 2026, and for most B2B founders it will through 2027 too. The honest version is:

Current priority stays Google. But AI discoverability is growing faster than Google traffic, and almost nobody is measuring it. So the smart bet is to measure it now, optimise for both, and position yourself before the shift fully happens.

Think of it like this. Google SEO is your existing revenue. LLM visibility is the new product line you haven't launched yet. Both get attention. Neither becomes zero.

The founders who figure out LLM visibility in 2026 will have a 12-24 month lead when the traffic mix tips. That's the prize.

A founder asked in the chat: "I'm launching a new site next week. One SEO thing to get right from day one?"

My answer: get your metadata right. Every single page needs a specific title, description, and OG image. Not "homepage - company name." Specific to the content of that page. Claude can generate these in seconds.

Everything else is a distant second. Rankings, backlinks, schema, programmatic pages, LLM visibility: they all compound once the basics are in. They mean nothing if your pages show up in search results with missing or generic metadata.

The reason this matters 10x: metadata is the only thing Google actually reads *before* it decides whether to show you. Everything else is optimisation within the window metadata already created.

Then: think about the structure of the page. Headings that match how people search. Content that answers the exact question. Internal links to related pages. Schema that tells Google what kind of content it is.

That's it. If you do those two things from day one, you're further ahead than 90% of new sites.