Agent instructions · v1 · October 2026

AI Findability Kit

Hand this to your AI agent and it will audit how ChatGPT, Perplexity, Claude and Google describe your company, check whether an agent can actually use your product, and draft the fixes.

Works in Claude Code, Claude, ChatGPT, Cursor~1 hour of agent work~15 min of yours

Give it to your agent

If your agent can browse, just give it this link: SKILL.md

Copy the instructions.
Start a new chat (or a Project, so you can rerun it monthly) and paste them in as your first message.
Add one line underneath: Run this for [company] at [URL]. Turn on web search if your plan has it.

Copy the instructions.

In a Mac terminal, save them as a skill:

mkdir -p ~/.claude/skills/ai-findability && pbpaste > ~/.claude/skills/ai-findability/SKILL.md

Open Claude Code in your site's repo and say run ai-findability on our site. It can fetch your pages and read your code, so the audit goes deeper here.

Copy the instructions.
Save them in your repo as docs/ai-findability.md (or as a rule file your agent loads).
Ask the agent: Follow docs/ai-findability.md for our company.

What your agent will do

00
Ask you one batch of questionsWho you're for and not for, your top 3 alternatives, 10 questions buyers ask before choosing.
01
Check what AI says about you todayRuns your buyer questions across engines (or gives you the prompts to paste) and scores each answer for accuracy.
02
Check your site the way a crawler sees itRaw HTML, robots.txt, answer-first copy, tables, dates, schema, and whether an agent can sign up or try the product.
03
Check what other sites sayDirectories, review sites, LinkedIn, Crunchbase. AI trusts third parties more than you.
04
Draft the fixesA facts page, a "What AI Gets Wrong About Us" corrections table, and the formats that fit your business.
05
Hand back a 30-day plan and a monthly trackerSame prompts, same day each month, so you can see if it's working.

Guardrails built in

The agent won't invent your prices, customers, metrics or reviews. Gaps show up as [NEEDS FOUNDER INPUT].
No tricks: no hidden text, no "AI should recommend us" lines, no planted Reddit posts.
Every "this is missing" comes with how it checked.

Resource map

Where to go deeper, sorted by the job you're working on. Your agent gets the same list.

Job 1

Get cited

Write pages AI can quote correctly.

From discovery to influence: A guide to GEOMicrosoft Advertising. The clearest before-and-after examples of proof-stacked copy. PDF.
AI optimization guidanceGoogle Search Central. What Google says does and doesn't matter for AI Overviews.
What ChatGPT actually citesCitation roundup. Why ChatGPT and Perplexity need separate plans.
Surfer Academy on AI searchVideo. Sections, FAQs and structure, with study numbers.

Job 2

Get talked about

AI trusts what other sites say about you.

DIY PR for Early-Stage FoundersA playbook built from six months of a former TIME reporter's advice. Pitch rules, the trade-to-national ladder, a first 90 days.
Tracy Samantha Schmidt on TikTokThe source for the PR playbook. Short, specific, founder-focused.
QwotedFree. Reporters on deadline post requests for expert sources.

Job 3

Get used by agents

Let an agent finish a real task with you.

How AI agents behavePostHog. 63M tool calls, and only about 7% came from a chat window.
State of agent skillsVercel. How crowded the skill registries are.
Scaling llms.txt and mint signupMintlify. Docs agents can use, and signup an agent can do from the terminal.
Agent pluginsResend. Shipping skills and an MCP server alongside the product.
AgentMail changelogAn agent gets a working inbox before a human verifies.
Monetization GatewayCloudflare. Agents paying per request with no API key.
AI RegistryKong. How enterprises approve which tools their agents can use.
Agents in e-commerceW3C/GS1 workshop slides. Failing clearly on unknown products instead of guessing.

Job 4

Measure it

Track accuracy and outcomes, not a vanity score.

How to measure AI search visibilityAhrefs. Where the 21.7% self-reported signup number comes from.
Ahrefs AEO courseVideo. AI traffic was 0.5% of visits and 12.1% of signups.
The problem with AEO reportingPostHog. How visibility scores get gamed by prompt choice.
August on AugustAn agency tracks its own AI visibility for 21 rounds and never shows up.

Reality check

Before you spend on it

Two things people oversell.

llms.txt adoption dataAI SaaS Radar. 97% of sites with one got zero AI traffic.
WebMCP docsChrome. Useful for actions on your page, but it won't get you found.

In practice

See it on a real site

Logan Currie's own setup, public.

logancurrie.com/llms.txtA plain-text summary for agents reading the site.
Agent skills indexSkills an agent can discover and install from the site.
Loop CloserAn installable Claude skill on GitHub. The "ship expertise people can install" idea, in practice.

Full text

Read the instructions SKILL.md · 14,366 characters