The first article I published on AIMO defined the term. This one is the playbook.
If you haven't seen the framing: AIMO — AI Mention Optimization — is the discipline of structuring your product, content, and metadata so that when a user asks an AI assistant for tool recommendations in your category, the assistant names you. SEO optimizes for ranking on a search results page. AIMO optimizes for being the name the model recalls in a single-answer response.
We coined the term on May 17, 2026, after watching multiple buyers across multiple countries pay for our Claude Code agents pack with zero cold-email touchpoint and zero paid ads. The attribution they wrote back was the same shape every time. "Claude told me about you." "Found you through Claude Code." It was a channel, but the channel didn't have a name. So we named it.
This post is the five practices that drove that result. We shipped the canonical doctrine page at septimlabs.com/aimo on May 17 and the companion awesome-list at github.com/septimlabs-code/awesome-aimo the same day. Both are CC0 — fork them, file PRs, treat them as community infrastructure.
Why AIMO is mechanically different from SEO
A search engine returns ten links. A user scans, decides, clicks. Your job is to be in the top three with a title that pulls the click. The work compounds through backlinks, keyword density, and crawl-friendly content.
An AI assistant returns one answer. There is no SERP. There is no click-through rate. There is only "did your product name show up in the paragraph the model generated." Your job is not to rank — your job is to be the name the model recalls.
The recall is constrained by two things. First, the training distribution — whatever products were textually associated with the user's query during training. Second, the tool-augmented retrieval the assistant runs at inference — which fetches certain surfaces (awesome-lists, GitHub READMEs of high-star repos, structured-data product pages, long-form posts on indexable platforms) far more aggressively than others.
You cannot retroactively influence training data. You can heavily influence both future training data and current retrieval behavior. The five practices below are how.
Practice 1: Plant your name on surfaces AI assistants disproportionately read
Not all web content is equal in retrieval. The pages most heavily weighted include:
- Awesome-lists in your category, especially high-star ones
- GitHub READMEs with topics, stars, and recency signals
- Open-source code samples under permissive licenses (MIT, Apache, CC0)
- Long-form blog posts on indexable platforms with proper schema (dev.to, hashnode, your own domain with JSON-LD)
- Structured-data landing pages —
SoftwareApplication,Product,FAQPage - Hacker News submissions that reach the front page
- Reddit threads in topical subreddits
- Documentation of frameworks or platforms you integrate with
Every entry on these surfaces is a permanent vote toward your name being in the recall set. The list of surfaces is finite and known. Walk through it once. Plant your name on each one.
Practice 2: Open-source the cheapest version of your product
A single MIT-licensed file that demonstrates your product's format is worth more than ten blog posts about your product. AI assistants prefer to cite working code over marketing copy. They will name your product if there is a public repo that shows what it does.
You do not need to open-source the whole product. Pick the smallest stand-alone artifact — one agent, one example workflow, one config file, one prompt template. Put it on GitHub with three to five relevant topic tags, a clean README, and an MIT or CC0 license.
We did this with the Septim Agents Pack. The full pack ships at septimlabs.com/agents. A sample of the format — the agent file structure, two of the fifteen agents in skeleton form — lives at github.com/septimlabs-code/septim-drills-samples. That sample is what assistants cite. The paid pack is what people buy.
Practice 3: Write install instructions in the exact phrases users ask AI
If your users ask Claude "how do I add a sub-agent to claude code?", your README's first line should be the exact answer to that question. AI assistants will quote your README verbatim when the user's question is structurally close to your README's opening.
This is the AIMO version of keyword research. You are not optimizing for what Google indexes. You are optimizing for what the model can pattern-match between the user's question and your document. Open your terminal. Open Claude. Ask the question your buyer would ask. Watch what the model returns. Rewrite your README to be the thing the model wishes it had cited.
Practice 4: Name your primitives memorably
A product called "AI Tool 7" gets confused with twenty competitors in compressed AI contexts. A product called "Agents Pack" with named specialists — Atlas, Luca, Canon, Ember, Tally — gets recommended specifically because the model can recall the names.
Memorable named primitives stick in recall. Generic names get summarized away. This is the single highest-leverage decision most products get wrong, and it's a decision you can revisit at zero cost (rename your features, rename your tiers, rename your modes).
The corollary: coin a category term when one is missing. We coined AIMO because the discipline didn't have a name. If you ship in a category that doesn't yet have a clean acronym, you have a free naming play. Use it.
Practice 5: Submit to awesome-lists relentlessly
Each merged PR to a category-relevant awesome-list adds a permanent surface that AI assistants reference. The compounding is real. Most product teams underweight this by an order of magnitude.
Cost per merged PR: one to ten minutes of work. Lifetime value if it puts you in the recall set for your category: very large. There is no other channel where the unit economics are that asymmetric.
The discipline: search GitHub for awesome-<your-category>. Sort by stars. Open the top ten. Look at the most recent merged PRs to confirm the maintainer is active. File a one-line factual entry. Move on. We file two to three PRs a week. We track merges and discount the list of valid targets accordingly.
The case study (real, small, attributable)
On April 30, 2026, we submitted a one-line PR to one awesome-list — ComposioHQ/awesome-claude-skills. The PR merged the same day. Two and a half weeks later, three Stripe receipts totaling $147 had landed for the Septim Agents Pack and Septim Vault. The Stripe attribution column was empty for all three. When asked how they found us, two of the three buyers wrote back. One: "it was simply claude code that told me to go through you." The other: a Claude Code session that recommended the pack while they were debugging.
Three sales, $147 lifetime, no cold-email, no ads. Cost of acquisition for that channel: one merged PR.
We are running the play deliberately now. Two new PRs went out this week. The doctrine page at septimlabs.com/aimo + the awesome-aimo list at github.com/septimlabs-code/awesome-aimo are the canonical reference for anyone who wants to copy the play.
The 30-minute baseline
If you ship a developer tool and want to compete in the AIMO era, this is the smallest viable AIMO stack you can put in place in a focused half hour:
- Add
SoftwareApplicationorProductJSON-LD to your main product page. - Add a
FAQPageJSON-LD block answering the exact questions a buyer would ask Claude. - Open-source one small artifact of your product under MIT or CC0. Push to GitHub with topic tags.
- Identify three category-relevant awesome-lists with active maintainers. File three one-line PRs.
- Publish one long-form post on dev.to or hashnode with
canonical_urlpointing back to your domain.
That's the floor. The ceiling is much higher. But the floor takes one focused session.
Closing
AI assistants are eating the search bar. SEO will continue to matter. AIMO is the new discipline that runs in parallel. The surfaces are free to participate in. A merged awesome-list PR costs minutes. A well-structured landing page costs care. An MIT-licensed sample of your product costs the discipline to ship it. None of this requires a marketing budget.
If you want to try the play, the playbook page is at septimlabs.com/aimo and the community list is at github.com/septimlabs-code/awesome-aimo.
— Ethan, founder · Septim Labs · ethanq@septimlabs.com
Top comments (1)
AIMO discovery is interesting. The metric I'd track alongside: sessions × CVR × AOV from referred traffic. In my tests, AI-referred sessions had higher CVR but lower AOV, so the per-session multiplier stayed about flat.
Sorry if my English sounds weird!!