P.S. If you want more AI-search and content workflows like this delivered to your inbox every week, join 6,000+ founders getting them free: wuki.beehiiv.com/subscribe
Three months ago I checked my analytics and found something I didn't know how to read: ChatGPT was sending me referrals. Except the referral page was /.
Not a blog post. Not a landing page. The homepage.
Twenty times in a single week, the same pattern — someone asked Claude or ChatGPT a question about solo-founder content strategy, the assistant answered with my name and article ideas, and the person clicked through to my homepage to find the post themselves.
Nineteen citations. Zero article clicks. That was the moment I stopped optimizing for ranking and started optimizing for being mentioned.
Here's what I learned in the five weeks since, and the framework I now run every old post through.
The Quick Answer That Kills Three
Google figured this out years ago. You look up "how to X", get the boxed answer, and never click.
AI search is that on steroids. The difference is that an LLM doesn't stop at quoting you once — it weaves your post's answer into its reply to a completely different follow-up three turns later. One good fact from your article can travel into seven separate conversations.
But here's the trap: the stats you see don't tell you that, because nobody clicks.
The first thing I did was stop using clicks as my success metric for AI search. Traffic got replaced by two things:
- Citation appearances in AI answers (searched manually, and increasingly via tools)
- Prompt queries reaching my content — sessions where an AI was visibly answering on my behalf
When I audited my articles that did get cited, a pattern emerged. Every one of them answered a question in the first 120 words.
The Audit I Want to Show You
The boring parts of my old posts were wasting all the potential. Here's what cheap little test: paste a blog post title into ChatGPT and ask "summarize this article in 3 bullet points worth quoting."
If it says "no information found on that" — or worse, hallucinates a generic summary with nothing specific is missing — the post isn't in the AI's usable index, or it's not structured in a way that survives distillation.
Underperforming posts shared three holes:
1. No obvious thesis sentence. My intros were scene-setting and meandering. AI cites what it can lift and reuse in two seconds.
2. Facts hiding in paragraphs instead of structure. A table or a stat in its own line gets pulled into answers 10x more often, because it's self-contained.
3. No "voice anchor." When an LLM decides whether to quote me or genericblog.com, it looks at the specificity of the author. Vague, carefully-written generic advice reads like everyone else.
The 90-Minute Rewrite Loop
I set up a simple pass for every post that had ever gotten organic traffic. It takes about 90 minutes per old post:
Pass 1 — The Answer Block (15 min). Rewrite the first 120 words as a standalone answer to the post's primary question. It should be quotable even if the rest of the article is stripped away. Title + first paragraph should alone be clear enough to answer: "Who is this for, and what will they learn?"
Pass 2 — Evidence Extraction (30 min). Convert any useful number or study into its own indented bullet or a one-line bolded statement. Trim every sentence that could apply to any SaaS ever.
Pass 3 — The Comparison Table (25 min). Whenever the post compares two things — manual vs automated, agency vs tool, do-it-yourself vs buy — a markdown table forces AI to lift exact rows into answers.
Pass 4 — Freshness Date (20 min). Update the publish date, and add a one-line "last verified" note near the top. AIs weight recency — a post that says 2024 data gets treated as stale, even when it's still right.
The Tool That Made This Repeatable
I won't pretend I keep doing this manually at scale. Once the framework worked on 5 posts, I needed it baked into my workflow — thinking about AI answers before the post exists, not after.
That draft pipeline is what I built at nextblog.ai — the posts get drafted with the answer-block structure in the first write where the bones of the article are laid down, SEO structure and potential citations in mind, instead of the scrappy old me rewriting everything three weeks after publishing for human-only writing. It's the same framework above, but moved earlier in the pipeline so the rewrite last month was the last one I needed.
What Actually Changed
Six weeks in:
- The articles I rewrote attract citations at a noticeably higher clip than the ones I didn't
- Referral volume from AI routing (people landing on a specific post) picked up
- I changed my definition of content ROI. A seat in the answer is now a legitimate objective, because a citation converts inside the AI conversation where debate can happen
The from-here-on-writing for humans is done. Writing that is also quotable by machines is the entire game now.
What's your take — do you chase clicks or citations?
Matthew — building the AI-native content routine that's existence-proof. nextblog.ai drafts your posts answer-first. If this was useful, join the newsletter.
Top comments (0)