Evaluation design. Provider claims are attributed to the sources below. This page does not claim a new independent DARC benchmark.
For: Developers and teams putting AI into practical use
⚡ TL;DR — the 30-second version
- Separate citations, mentions and visits.
- Record date, engine, query and evidence for each observation.
- Do not publish an uplift without a real comparison and sample.
Choose a fixed query set and record engine, date, region, account state and the complete answer. Save the cited URL, the claim it is attached to and whether the cited page actually supports that claim. A brand mention without a link should not count as a cited source.
Repeat observations and keep a control group. AI answers can change across runs; a single favorable answer is not evidence of a stable traffic gain. Inspect referral analytics separately from answer screenshots.
“A citation, a mention and a visit are three different events. Measure them separately before telling a growth story.”
Make pages useful and inspectable: direct answers, current facts, source links and honest update dates. This site's new model catalog follows that principle by distinguishing publisher announcements from independently replicated tests.
Start with measurement consistency before making a growth claim. Publish the number of queries, observation frequency and the rate of ambiguous citations; retain enough evidence for another person to audit the classification.
Field guide / 01
Define the observation before collecting it
Separate brand mention, linked citation, claim-supporting citation and actual referral. These are different outcomes and should never share one success column. Record engine, query, date, region and account conditions for each observation.
Use a fixed query set with several intents rather than prompts constructed to force a mention. Repeat the set and preserve raw answers under the applicable service terms.
Field guide / 02
Do not infer causality from a screenshot
An answer changing after a page edit does not prove the edit caused the change. Engines, indexes and query interpretation can change independently. Use comparison pages or periods where possible and report the limitations.
The first useful deliverable is a consistent evidence ledger, not an uplift percentage. Once the measurement is stable, use it to prioritize genuinely better pages and original evidence.
Before you ship
0/5Use this as a review aid. Checks reset when you leave this page.
Questions that come up
Does a citation imply a visit?
No. Measure citations in the answer and referrals in analytics separately. A cited source may receive no click, and a visit can arrive through another path.
Sources & verification
Last checked 23 Sept 2026. Publication dates and access conditions belong to the linked originals.
- 01OpenAI API changelogOpenAI↗
- 02Gemini API release notesGoogle↗
- 03Perplexity — Sonar and Search API changelogPerplexity↗