Optimly AI Index | AI-Agent Retrieval Benchmark
Weekly AI retrieval benchmark
2026-W37 · 2026-09-06 – 2026-09-13
This week’s read
User-directed AI demand rose this week.
Brand-page activity rose 47.9%. Internet Registration Data Services had the most observed activity.
Direction: Rising
Confidence: Directional
Main driver: User-directed activity
All automated activity
485,298
↑ 69.3% requests this week
Only the user-directed subset is a proxy for demand. This total also includes search crawling, model training and activity with no clear purpose.
What AI agents were retrieving this week.
Leading research themes
Ranked by observed topic-page requests
Top five = 48% of topic activity
- How to Measure AI Referral Traffic and Customers
Solution research · Unattributed AI agent
17 page requests - AI Search Optimization for B2B SaaS
Solution research · Unattributed AI agent
15 page requests - Conservative Alternatives to Aarp
Evaluation criteria · Unattributed AI agent
New this week
14 page requests - AI Brand Visibility vs Accuracy: What Should You Measure?
Evaluation criteria · Unattributed AI agent
13 page requests - Parks and Recreation Management
Solution research · Unattributed AI agent
New this week
9 page requests
How to read the signal
The panel concentrated on a small set of related research questions.
48% of observed topic activity went to the five leading themes. That concentration is more informative than the total request count, but the sample remains directional.
Research stage
Problem research
- Continuing to Wear Existing Wardrobe Without Purchasing New Items
- Build An App Without Coding
Evaluation criteria
- Conservative Alternatives to Aarp
- AI Brand Visibility vs Accuracy: What Should You Measure?
Solution research
- How to Measure AI Referral Traffic and Customers
- AI Search Optimization for B2B SaaS
Topic × agent observations
These are direct path-level observations. Multiple requests can come from the same agent and do not represent distinct people.
- padel rackets for power
18 requests · PerplexityBot (18) - enterprise computer vision api
10 requests · PerplexityBot (6), ShapBot (4) - enterprise generative ai
8 requests · PerplexityBot (8) - enterprise asset management software reviews
7 requests · PerplexityBot (6), Unattributed AI agent (1)
Unmet agent demand
Repeated failed or partially failed paths form a practical research queue: a new profile or comparison may be missing, or an existing request may need a canonical redirect or taxonomy fix.
- Adobe Experience Manager Assets
Profile or routing opportunity
ChatGPT-User (15), PerplexityBot (1)
16 observed errors · 503 × 16
16 path requests
15 user-directed - Hacker News Show Hn
Profile or routing opportunity
ChatGPT-User (8), Bytespider (1)
9 observed errors · 503 × 9
9 path requests
8 user-directed
Panel size and reliability
User-directed visits: 35,745
↑ 559.3%
Best available proxy for someone actively using an AI product.
Brand-page visits: 24,479
↑ 47.9%
Visits to resolved brand profiles, including every agent purpose.
Buyer-topic visits: 4,374
↑ 706.7%
Observed visits to topic pages that group related buyer questions.
Known-page match rate: 68.8%
No comparable baseline
Share of automated visits matched to a page the Index publishes.
Compared with the wider AI web.
This week's strongest findings
User-directed activity, as Cloudflare labels it, is 1.9× more concentrated in this panel than across Cloudflare's network 35,891 visits · 7.8%
Cloudflare labels 7.8% of the AI-agent visits to this site as "User Action", against 4.2% across its whole network."Automated, no company named" moved down from 88.5% to 75.4% of all visits 366,060 visits · 75.4%
This is one of Optimly's five groups, measured against every visit Cloudflare flagged as automated on optimly.ai during the week.32.3% of requests this week returned an error 147,867 visits · 32.3%
147,867 visits of 458,468 visits returned a status of 400 or above.
What buyers appear to be asking
15 question topics touched
Distinct topics that an AI system read at least one page about this week.
| Topic | Requests | Most active agent |
|---|---|---|
| How to Measure AI Referral Traffic and Customers | 17 | AI Crawler (behavioral) |
| AI Search Optimization for B2B SaaS | 15 | AI Crawler (behavioral) |
| Conservative Alternatives to Aarp | 14 | AI Assistant (behavioral) |
| AI Brand Visibility vs Accuracy: What Should You Measure? | 13 | AI Crawler (behavioral) |
| Parks and Recreation Management | 9 | AI Assistant (behavioral) |
| ... | ... | ... |
Which categories moved this week
| Category | Visits this week | Change |
|---|---|---|
| Emergency Medical Training & Certification | 13 visits this week · no comparable count for the week before · New this week | |
| Machine Learning Operations (MLOps) Platforms | 13 visits this week · no comparable count for the week before · New this week | |
| AI Agent Evaluation Platforms | 13 visits this week · no comparable count for the week before · New this week | |
| Healthcare Regulatory Compliance Software | 12 visits this week · no comparable count for the week before · New this week | |
| K-12 Online Learning Resource Platforms | 12 visits this week · no comparable count for the week before · New this week | |
| ... | ... | ... |
Most-read brand pages
| Brand | Category | Visits | The week before | Change | Most active agent |
|---|---|---|---|---|---|
| ICANN | Internet Registration Data Services | 16 | 38 | 58% below last week Small sample |
ChatGPT-User |
| ... | ... | ... | ... | ... | ... |
Methodology
Two parts, in order: what the report can and cannot tell you...