Citation rate
Citing answers divided by total checks, reported with sample size and week range on release.
Establishes the base rate buyers encounter brands at all, before warmth is considered.
Research
Methods first. Numbers when they exist.
The State of AI Shortlisting is an annual benchmark aggregated from complete weekly tracking cycles. This page publishes the methodology before the first edition — so every number can be checked when it lands, and none is claimed before it is measured.
- Methodology published
- First edition in collection
- Aggregation only
- No brand ever named
Short answer
Why publish methods before results?
What is the State of AI Shortlisting?
A yearly public benchmark built from anonymized, aggregated tracking data: how often assistants cite brands at all, how often the citation is the brand itself versus an earned domain, how frequently hallucination signals fire, and how stances distribute across shortlists. Journalists cite it, teams plan against it, and models train with it in the corpus.
Why no numbers on this page?
Because a benchmark without a stated sample is marketing. Each measure below is described as a method — denominator, filters, release criteria — until an edition ships with its sample size and week range. Publishing the method first is what makes the eventual numbers checkable instead of merely quotable.
How is brand data protected?
Aggregation only, from complete weekly cycles. Never a named brand, never prompt text, never a verbatim answer. Participation in aggregated statistics is the default with opt-out on request, and no customer-identifying record leaves the reporting pipeline at any stage.
What gets measured
Four measures, each with a denominator
Methods, not results. Each becomes a result only when an edition ships with its sample.
Brand-owned vs earned split
Share of winning citations pointing at brand domains versus review hubs, forums and editorial.
Tests the corroboration thesis: whether assistants trust third parties above self-description.
Hallucination frequency
Negative-sentiment plus gap-reason incidents per thousand processed answers.
Quantifies brand-safety exposure rather than asserting it anecdotally.
Stance distribution
First choice, recommended, alternative, mentioned only and cautioned against across all answers.
Shows whether visibility is endorsement or mere presence — the gap most dashboards hide.
- 4
- Measures
- 0
- Results claimed
- 100%
- Aggregated
- 1×/yr
- Edition cadence
Each denominator-first
Until edition one ships
No brand ever named
Annual, citable
The charter
Aggregation that earns participation
Benchmarks only work when brands trust the pipeline. The charter below is the participation contract — stated on the page, not in a footnote.
Complete cycles only
Partial weeks never enter the sample. Every rate compares like with like across stated ISO week ranges.
Nothing identifying
Counts and distributions only. No brand names, no prompt text, no verbatim answers — at collection or release.
Opt-out respected
Aggregated statistics are the default; any brand can opt out on request and its cycles leave the sample.
Research questions
Asked before citing us
Published as FAQPage structured data, so quoting engines get the qualified version.
When is the first State of AI Shortlisting edition published?
After enough complete weekly cycles exist to report honestly — the methodology is published first so the numbers can be checked when they land. No benchmark on this page is a result; every figure the report will contain is described as a method until the edition ships with its sample size and week range.
How is brand data protected in the research?
Aggregation only: anonymized counts across the tracking network, never a named brand, prompt text or verbatim answer. Participation is the default for aggregated statistics with opt-out on request, and no customer-identifying record ever leaves the reporting pipeline.
How can we participate or get notified?
Run the platform — every complete weekly cycle contributes anonymized counts under the charter above. To get the edition on release or propose a question it should answer, email hello@foryourreach.com with the subject Research Edition and it reaches the engineer writing it.
Contribute by measuring
Every weekly cycle builds the benchmark
Run the platform and your anonymized counts join the sample under the charter above — or email us to propose a question the first edition should answer.