how to get your tool recommended by LLMs

From never being mentioned by AI, to being recommended as the #1 tool. Here's the playbook.

Hey,

a few months back I noticed something that annoyed me more than it should have: my main product, the one that gets reviewed as #1 in its category, had quietly stopped showing up in AI recommendations.

Ask ChatGPT for the best tool in the space and it named two competitors I’d barely heard of. Both were a few months old. Both had blogs with thousands of pages, each one built around the exact phrases people type into a chat box. They hadn’t out-built me. They’d out-published me.

So I spent a quarter running the blueprint below to take the spot back. It worked. Every step is linked, most have a time estimate, and the whole thing is free.

In today’s email: how assistants actually pick recommendations, two hours of plumbing for week one, the 100-page seed hub, the off-site placements that matter more than your blog, and the weekly loop your coding agent runs while you build.

How LLMs pick recommendations, in 60 seconds

• Every AI recommendation is a web search plus a summary. ChatGPT searches Bing’s index. Claude searches Brave: Anthropic’s subprocessor list still names Brave in mid-2026, and almost nobody optimizes for it. Perplexity crawls its own index. Google’s AI answers fan one question out into many sub-searches, so a page per subquestion is how you get pulled in.

• Models shortlist brands from training memory first, then hunt for sources to justify the pick. Familiarity compounds, which is why the off-site work below counts as much as your own blog.

• GPTBot, ClaudeBot and PerplexityBot don’t execute JavaScript. Vercel’s network data shows they download JS files and never run them. Client-rendered content is invisible to every assistant at once.

That’s the theory. Everything below is the blueprint.

Week 1: two hours of plumbing

Five jobs, in order. None needs a developer.

1. Verify your site in Bing Webmaster Tools (15 min). It’s the front door to ChatGPT and Copilot visibility, Microsoft says so directly, and almost no indie founder has done it.

2. Turn on IndexNow (10 min). New pages get pushed to Bing, Yandex and Naver instantly instead of waiting on a crawl. On Cloudflare it’s one free toggle; on WordPress it’s free in Rank Math. Google ignores IndexNow; an honest sitemap lastmod covers Google.

3. Open your robots.txt and check who you’re blocking (10 min). OpenAI runs separate bots: you can block GPTBot (training) and keep OAI-SearchBot (citations). The popular one-click “block AI bots” settings kill both. I’ve watched people opt out of being recommended and call it defense.

4. Curl your key pages (5 min). If your product description isn’t in the raw HTML, no assistant can read it. Fix with server-side rendering before you write a word of content.

5. Claim your profiles on G2, Capterra and Product Hunt (30 min). 34.5% of AI Overviews for commercial queries cite a review platform, and the basic listings are free.

Weeks 2-4: seed the first 100 pages

1. Mine the exact phrases buyers type (one evening). Google autocomplete with a-z suffixes and People Also Ask (keywordspeopleuse.com mines both, free trial), plus Reddit threads in your niche for the literal sentences people paste into chat boxes. Bank 300+ phrases. Each one is a page.

2. Sort them into five proven shapes. “Best X for Y”, “X vs Y”, “X alternatives”, “how to do X in Y”, and one page per integration. Zapier’s programmatic playbook is the archetype: profile pages for head terms, pair pages for the long tail.

3. Give every page the citability treatment. One sourced statistic, one quotable line, cited references. The Princeton GEO paper measured up to 40% higher AI visibility from exactly these additions, keyword stuffing measured negative, and lower-ranked pages gained the most. The per-page checklist version is in today’s kit.

4. Draft with AI, gate with a human. Google’s scaled content abuse policy doesn’t care who wrote the page. It cares whether the page helps users or exists to game rankings. Every page answers a real question or it doesn’t ship.

Months 2-3: drip, place, iterate

• Publish 25-30 pages a week, never 500 in a day. That cadence comes from the cleanest indie case study I found (512 pages, 11,840 clicks/month). Crawl budget is a non-issue below about a million pages; pacing is about quality signals, not crawler mercy.

• Pitch, pay, or copy your way into 10 third-party listicles. “Best X” roundups are 43.8% of everything ChatGPT cites for those prompts, and freshness beats authority: 79% of cited lists were updated within the year, a third sit on small domains. Offer authors your real numbers so their post gets fresher; updated-this-month is what gets cited.

• Spend PR effort on mentions, not backlinks. Branded web mentions correlate 3x more strongly with AI visibility than backlinks (0.664 vs 0.218). Podcast appearances, directories and “tools I use” posts all count now.

• Post where your engine eats, never on one surface only. Semrush’s 230K-prompt study: Perplexity leans Reddit, Google’s AI Mode leans LinkedIn, and ChatGPT’s Reddit citations collapsed from roughly 60% of responses to 10% in six weeks. Reddit still pays twice, since its content is licensed to both Google and OpenAI.

• Know the line. NationalToday published 850,000 AI pages, got the directory manually nuked, and its ChatGPT citations vanished with it, because assistants lean on the same indexes.

A Google penalty doesn’t just kill your rankings anymore. It kills your ChatGPT citations too.

Skip these (verified time-wasters)

• llms.txt. Ahrefs checked 137,210 domains: 97% of the files got zero traffic and no AI bot ever requested a missing one. Google’s John Mueller compares it to the keywords meta tag, which should tell you everything. Skip unless you ship developer docs.

• Special “AI schema.” Google’s own docs: no extra markup gets you into AI answers. Anyone selling AI-Overview schema is selling air.

• Google Indexing API tricks. Restricted to job postings and livestreams, with a spam crackdown on the loopholes. IndexNow plus honest lastmod is the legitimate version.

The weekly loop: your coding agent runs it

1. Wire Search Console into Claude Code with mcp-gsc (10 min with a service account; open source, 1.3K stars). Bing has an MCP server too.

2. Run the striking-distance prompt weekly (it’s the tip of the week, below).

3. Segment AI traffic in GA4 (15 min). This walkthrough builds a custom channel group on the regex chatgpt.com|perplexity.ai|claude.ai|gemini.google.com|copilot.microsoft.com, and it applies retroactively. Expect tiny numbers with outsized value: Ahrefs measured AI search at 0.5% of their traffic but 12.1% of their signups.

4. Audit the assistants monthly (30 min). Ask ChatGPT, Claude and Perplexity your 10 money prompts and log who gets named, because only 49% of brands stay consistently visible over three weeks. Otterly tracks 15 prompts daily for $29/month, Gego is the open-source version you run on API credits, and the manual audit script in today’s kit costs $0.

Vibe coding tip of the week

Give Claude Code your Search Console and ask it this, weekly: “Using GSC data from the last 28 days, list pages ranking 8-20 with over 100 impressions. For each, tell me which query it almost ranks for, then rewrite the intro to answer that query directly with one cited statistic and one expert quotation.” That prompt is the highest-ROI recurring task I’ve given an agent this year.

What we liked from around the web

• Google users click results on 8% of searches with an AI Overview, vs 15% without: Pew’s study is the cleanest evidence the old click economy is shrinking.

• ChatGPT now handles about 12% of Google’s search volume but sends 190x less traffic: the whole tension of this issue in two numbers.

• AI referrals to top sites grew 357% in a year: still a rounding error next to Google, but nothing else on your traffic report grows like this.

• One study claims AI assistants now equal 56% of global search volume: Rand Fishkin disputes the math, and the fight itself tells you how contested this ground is.

• Profound’s analysis of 680M citations: the deepest public dataset on what each AI platform actually cites, if today’s issue left you wanting the full tables.

One thing before you go

Grab the SEO/GEO Engine kit: the per-page GEO checklist from weeks 2-4, the 4-prompt content engine that goes from seed keyword to linked post, and the monthly 10-prompt audit. Free, built to be handed straight to your agent. Do the week-one plumbing today; the kit picks up from there.

Cheers,

Jonathan