Observed AI answer
The exact completed provider response, buyer question, model, timestamp and returned source URLs. A missing or failed response is never replaced with an estimate.
Eli research
Research on how people and AI choose what to buy
Benchmarks, experiments and open data from Eli. We publish what the evidence shows, including results that do not support the story we hoped to tell.
Start with the data
View all open dataMethod
Agent Revenue Index
How Eli will test whether AI can find, verify, compare, price and act with B2B software companies.
Live indexAI Visibility Index
Repeated buyer-question measurements across categories, with sample counts and confidence ranges.
MethodHow the index is measured
Question selection, providers, failure handling and what the numbers do not prove.
ReleasesImmutable release pages
Every published result stays inspectable in the release it came from.
Open dataDownloadable datasets
Public datasets and publication records you can inspect, cite or build on.
PressPress room
Company facts, research summaries and approved media assets.
Study index
Which studies have published data?
Every study is listed with its status. Where observations have been released, the sample, measurement window and assistants are shown, and the raw data is linked where it is released. Where only the method is public, no numbers are shown.
| Study | Status | Sample | Window | Assistants | Data |
|---|---|---|---|---|---|
| AI search by buyer stage | Method published, data pending | Not released | Not released | Not released | Not released |
| AI search by B2B category | Method published, data pending | Not released | Not released | Not released | Not released |
| Study | Status | Sample | Window | Assistants | Data |
|---|---|---|---|---|---|
| AI Visibility Index | Data published | Per category; sample counts and confidence ranges are shown on each release | Rolling releases | Listed per release | Open data |
| Recommendation rate benchmark | Method published, data pending | Not released | Not released | Not released | Not released |
| ChatGPT, Perplexity, Gemini and Claude compared | Method published, data pending | Not released | Not released | Not released | Not released |
| Study | Status | Sample | Window | Assistants | Data |
|---|---|---|---|---|---|
| Citation benchmark | Results published, raw data pending | 431 answers: 366 ChatGPT, 65 Gemini | 16 September to 15 October 2026 | ChatGPT, Gemini | Not released |
| Who ChatGPT and Claude recommend as AI marketing agents | Results published, raw data pending | 24 answers to 12 buyer questions a week | Weekly since 20 October 2026 | ChatGPT, Claude | Not released |
| AI search statistics | Results published, raw data pending | 431 answers: 366 ChatGPT, 65 Gemini; each figure gives its own count | 16 September to 15 October 2026 | ChatGPT, Gemini | Not released |
| Study | Status | Sample | Window | Assistants | Data |
|---|---|---|---|---|---|
| AI search to revenue benchmark | Method published, data pending | Not released | Not released | Not released | Not released |
| Study | Status | Sample | Window | Assistants | Data |
|---|---|---|---|---|---|
| Agent Revenue Index | Method published, data pending | Not released | Not released | Not released | Not released |
| Agent-ready website index | Method published, data pending | Not released | Not released | Not released | Not released |
| AI agent action readiness benchmark | Method published, data pending | Not released | Not released | Not released | Not released |
Methods and change history: research methodology, benchmark methodology and index methodology. Last reviewed 19 October 2026.
How the research is used
Where do these methods show up in the product?
The same rules that govern this research govern what Eli tells a customer: repeated measurements instead of a single run, saved answers and sources behind every number, and observed results kept apart from claims about cause. See how AI search works for B2B buying, what an AI search execution platform does with a measured gap and how Eli compares with other platforms.
Evidence standards
The evidence hierarchy
These standards define what Eli records, what it may infer and what must remain explicitly unproven.
Verified public work
A public URL that Eli can fetch after publication, with a receipt for the approved action and a comparable follow-up check.
Buyer-reported influence
A buyer says an AI assistant influenced the journey. This is useful attribution evidence, but it remains separate from verified revenue.
CRM-verified outcome
A consented lead, opportunity or closed outcome matched to the company record. This is the strongest commercial evidence in Eli's learning loop.
Open original data
Inspect the release, not only the chart.
What is public
Eli's public AI Visibility Index exposes its repeated buyer questions, measured providers, sample counts, failure handling and confidence intervals. Immutable release pages and downloadable files make each published result inspectable. Eli does not claim a billion data points or hide an unavailable provider behind a modeled score.
Publishing rules
How Eli publishes research
A public result is useful only when another person can understand where it came from and what it does not prove.
Sources
Factual vendor, platform and policy claims should link to current primary sources. Eli's own measurements link to the published method and underlying release where available.
Dates
Research pages show a publication or review date when freshness materially affects the claim. A date changes only when the content or underlying data changes.
Authorship
Eli is identified as the organizational publisher. A named human reviewer is shown only when that person materially reviews the publication.
Automation
Agents may research, draft and prepare work. Approval, publication authority and third-party outreach remain governed by the workspace controls and the connected system.
Comparisons
Competitor pages separate current vendor claims from Eli's interpretation and tell buyers what to verify. Eli does not invent customer reviews or awards.
Corrections
Material factual errors should be corrected in the public page and its saved evidence. A failed read is shown as unavailable rather than replaced by a plausible value.
What Eli may optimize
- Priority among supported buyer questions
- Owned pages and approved website journeys
- Human-final outreach to relevant third-party sources
- The next experiment after comparable follow-up evidence
What Eli does not claim
- A guaranteed ranking or AI recommendation
- Causality from a before-and-after change alone
- A brand mention on a page that was not verified
- Revenue without a buyer report or CRM-confirmed outcome
Eli on Eli · August 2026
The publication sprint is a measured experiment, not a ranking claim
Eli published a connected resource portfolio to test whether buyer-decision coverage, internal distribution and verifiable public pages improve discovery. The current release contains 122 canonical resource pages, eight topic hubs and a worksheet for every guide. A quality check found no exact repeated editorial blocks and no suspiciously similar page pairs. That proves the portfolio is public and structurally distinct; it does not prove that a search engine indexed, cited or ranked it.
122canonical resource pages
Each URL owns a distinct stated intent and canonical.
8topic hubs
Every resource is connected to a navigable commercial cluster.
197URLs sent through IndexNow
Accepted submissions are discovery signals, not index guarantees.
3Google priority inspections
The strongest commercial pages were manually queued for recrawl.
The seven receipts we keep separate
- 01Published
The canonical URL returns the intended public page.
- 02Discovered
A sitemap, internal link or crawler request exposes the URL.
- 03Crawled
A known search or AI crawler requests the page.
- 04Indexed
A search engine confirms the URL is in its index.
- 05Cited
A saved AI answer returns the exact page as a source.
- 06Visited
Permitted analytics records a referral or landing-page session.
- 07Commercial outcome
A consented action reaches a trusted CRM or billing record.
Release note: 20 August 2026. The counts above describe publication and submission receipts from this release. Eli will add crawl, index, citation and commercial results only when those separate observations exist.
Standard version 1.2 · Updated 20 August 2026 · Eli
About Eli · facts checked 20 October 2026
What Eli is: AI agents that do your marketing work
Eli is an AI growth team for B2B companies: four autonomous AI agents for AI search, your website, conversions and Meta ads. They work every day; you approve what matters. You do not do the marketing work yourself: Eli's agents do it every day, also when you are not logged in, and ask you only when a decision is yours.
- AI Search Agent: gets you named when buyers ask ChatGPT and Gemini what to buy.
- Website Agent: fixes the pages where buyers decide; changes go live after your yes.
- Conversion Agent: tracks which visits become customers, through the Eli site tag.
- Paid Ads Agent: tests Meta ads inside your budget, after a tracking check.
Eli is for founders of B2B software and service companies, often 11 to 50 people, whose website gets visitors but not enough customers, and who have no marketing team to work on it.
You use Eli from ChatGPT or Claude, through Eli's MCP connector, or in your Eli account at hireeli.io, and Eli sends you one short report a week. There is no new dashboard to learn.
Eli Free is free forever: one website, a diagnosis in the first 48 hours, then 5 answers once a week, no card. Eli Autopilot has all four agents (AI Search, Website, Conversion and Paid Ads for Meta ads, Meta ads by invitation while Meta reviews Eli) on every plan; only the capacity differs: Launch $99, Core $249 or Grow $599 a month (or €99, €249 and €599). Eli Done for You is $1,500 or €1,500 a month: Grow's capacity, set up and run with Eli's team. Every plan is monthly and can be cancelled anytime, and every account starts free. Enterprise is scoped and priced in a call. Your ad budget is separate and goes directly to Meta.
On G2, Eli is rated 4.8 out of 5 from 2 verified reviews (checked 20 October 2026). G2 reviews
You approve Eli's first change and your ad budget. After that you can let Eli improve pages and publish by itself, and you can undo anything. More money or anything new still waits for your yes. "Stop everything" works at any time. Eli never promises a ranking, a revenue figure, a multiple or a date. It tracks conversions, reports what it measured, and says so when it is too early to tell.
How Eli works · Plans and prices · Use Eli in ChatGPT or Claude
Inspect a real baseline on your own website
Free account. No card.
Run the free checker