Cited ≠ installed
ALG vs GEO: when the buyer is an agent, not a search box.
By Daria Dovzhikova · Updated October 2026
Generative engine optimization answers one question: on an unbranded buyer prompt, does ChatGPT, Claude, Gemini, or Perplexity name you?
Agent-led growth answers a different one: when a coding agent plans the stack, does it install you?
Both matter. Mixing them into one "AI visibility" score is how you celebrate a citation while the agent never finishes setup.
Two funnels
| Compared | GEO | Agent-led growth (ALG) |
|---|---|---|
| Buyer | Human asking an answer engine | Coding agent (Claude Code, Cursor, Codex) mid-build |
| Win | Named / cited in the answer | Installed and used in the session |
| Surface | Pages engines extract: comparisons, docs summaries, third-party mentions | Repo, skills, MCP, docs the agent opens |
| Failure mode | Rival cited on your own comparison page | Agent hits a wall and routes around you |
Anže Voje on the State of B2B GTM thread: many founders now procure mostly through Claude Code / Codex. Mariane Bekker: Agent Led Growth > Developer Led Growth. Matias Gontovnikas at Supabase: working on a lot more agent-led growth now.
That is not "SEO with a new acronym." It is a distribution channel where the first evaluator is software.
Where GEO still wins
GEO is still the right job when the buyer asks an answer engine before they open a repo. We measure it as four numbers per engine, branded recall on its own line — not a blended dashboard. See GEO services.
If engines already open your pages and misread them, fix those pages. That work does not disappear because agents also buy.
Where ALG starts
ALG starts when the agent is mid-build and needs a tool. The RouterGrowth demo (3 Oct) line lands: Claude Code can plan the whole go-to-market — then it hits a wall. The wall is usually docs, auth, or a missing skill, not a missing blog post.
Ship for that path:
- Install and first success in one agent session
- Skills / instruction files that complete one real task (see SKILL.md is the install path)
- Pricing and packaging the agent can reason about (work units, not only seats — covered on GTM for AI agents)
What we sell against this
Growth Pilot can carry both as experiment cards: one GEO baseline/re-measure, one AX / install-path card. Friday kill/keep. $5,000 fixed, four weeks. We do not sell "rank inside ChatGPT," and we do not sell a skill file as a silver bullet.
Skip the mash-up. Another GEO 101 will not teach an agent to install you. Another agent-hype post will not fix the comparison page engines already cite wrong.
Related: GEO services · AI Agents playbook · SKILL.md is the install path
Related guides
Want both cards run?
One GEO baseline and re-measure, one install-path card. Friday kill/keep. $5,000 fixed, four weeks.