Agents got what they needed 74% of the time (n=118).
One fix from last week is still inconclusive: only 6 calls reached the changed tool, too few to tell.
23%
of 600 AI answers cited Relaykit. That's your Share of Agent.
n=600 · 95% range 20–27%
About the same as last week (+1 pt, too small to call).
1 of 1 fix we re-checked worked.
Share of Agent
23%
About the same as last week (+1 pt, too small to call)Your true share very likely sits between 20–27% (95% range, n=600).
You win 5 of 30 buyer questions; Courierly wins 17 of 30. A question counts as won when assistants consistently cite you for it.
Share of the same answers that cited each name. Whiskers are 95% ranges.
n=600 answers · 5 of 5 assistants
Based on AI assistants' answers, which can be wrong. Check the cited sources.
Agents that tried your API or MCP server got what they needed 74% of the time.
n=118 · 95% 65–81%
Distinct agents at each step this week, and how many fell away between steps.
AI-bot requests, checked against each vendor's published ranges
Noise we removed312 removed, never counted
Two fixes this week, most important first.
Fix 1: Fix get_delivery_status for older messages
Three agents asked for the status of a message and got an empty result. They rarely retry.
Pair this with the code change: return a status for message ids older than 24 hours instead of an empty list.
Download fix 1 as a .patchWe re-check on Oct 26 and tell you whether it worked.
Fix 2: Add a code example to your idempotency page
Fix get_delivery_status for older messages
Only 6 calls reached this tool after the change, too few to judge. We will re-check next week.
verdict moves to next week's re-checkUnblock OAI-SearchBot
Refusals fell from 31% to 3% of 68 requests.
Your go-to-market plan for the next four weeks.
Buyer questions where someone else gets cited and you mostly don't.
“Which email API works with AI agents through an MCP server?”
You: 1 of 20 answers · Courierly leads 65%
Source fact: found.question.mcp_server.share“How do I make email sends idempotent so retries never double-send?”
You: 2 of 20 answers · Sendwright leads 55%
Source fact: found.question.idempotency.share“Which email API has the best documentation?”
You: 3 of 20 answers · Courierly leads 70%
Source fact: found.question.documentation.shareThird-party pages the assistants already cite on your questions. Earn a mention there.
Cited 14 times in answers that skipped you, by Perplexity, ChatGPT.
Example page: reddit.com/
Cited 11 times in answers that skipped you, by Claude, Gemini, Perplexity.
Example page: emailtoolsreview.example/
Cited 9 times in answers that skipped you, by Claude, ChatGPT.
Example page: registry.modelcontextprotocol.io/
3 done · 3 to do · 0 not checked. The ways agents read and use a product, checked against your site.
Docs as markdown
To doServe every docs page at /docs/<slug>.md and answer Accept: text/markdown.
Source fact: fix.readiness.markdown_docsMCP registry listing
To doList the MCP server in the official MCP servers registry with a one-line description of what agents can do.
Source fact: found.source.registry.modelcontextprotocol.io.citedAI search bots allowed
To doAllow OAI-SearchBot, Claude-SearchBot and PerplexityBot in robots.txt.
Source fact: fix.readiness.robots_ai_searchllms.txt
DoneOpenAPI file
DoneMCP server
DoneWeek 1Open the doors
Week 2Answer the three questions
Week 3Earn the mentions
Week 4Measure and repeat
Built from this week's 600 answers and your agent traffic with fixed rules, no guessing. Every line points to the number behind it. · method found-v2 · used-v4 · fix-v1
Latest agent activity6 events · last at Oct 18 · 21:14 UTC
Claude Code called search_docsgot data
got dataOAI-SearchBot fetched /docs/webhooksblocked (403)
blocked (403)Cursor called list_templateserror (500)
error (500)ChatGPT-User fetched /pricinggot data
got dataGitHub Copilot connected, listed toolsno call followed
no call followedClaude Code called get_delivery_statusempty result
empty result