Research · Engine Index

How the answer engines behave, measured — not modelled.

An aggregate over every completed audit in Synthic, refreshed hourly. Only averages leave the database, and a rate is printed once an engine has 25+ answers behind it.

Add your audits to the index

FIG.00 — every rate is an average over stored answers · n on every rowLIVE

160

verbatim answers

completed audits

brands audited

90d

rolling window

Finding 01

Perplexity shows its sources. Gemini does not.

Not a quality ranking — a difference in kind. One answers from a live retrieval pass and hands back what it read; the other answers from training. A mention you cannot trace to a source is a mention you cannot check, or influence by changing your site.

Perplexity

100%

of 76 answers carried a source

Gemini

of 84 answers carried a source

Finding 02

The warmest engine is the least grounded one.

Warmth is not accuracy. The engine most likely to speak well of a brand is the one least able to say where the impression came from — which is exactly the sentiment you cannot correct by fixing a page.

Gemini

93%

positive · 0% of its mentions traceable

Perplexity

80%

positive · 100% of its mentions traceable

Finding 03

Gemini takes 2.7× longer to answer than Perplexity.

Measured at the API, on the same prompts, over the same window. It is a fact about the providers' own endpoints, not about the network between you and them.

Perplexity

4.7s

mean round trip

Gemini

12.6s

mean round trip

Every engine, every measure

Engine Index · last 90 days160 answers · 16 audits · as of Nov 2

#	Engine	Names the brand	Cites a source	Grounded	Positive	Latency	n
01	PerplexityPerplexity AI	20%	100%	100%	80%	4.7s	76
02	GeminiGoogle	18%	0%	0%	93%	12.6s	84

Averages over stored answers from every Synthic workspace, 90-day window. A rate is shown once an engine has 25+ answers; until then it is collecting.

Method

What each column means.

Plain averages over stored rows. No weighting, no smoothing, no imputation for engines that don't cite.

Names the brand: Share of answers in which the engine mentioned the audited brand at all. Low numbers are normal: most brands are not yet in most answers.
Cites a source: Share of answers that carried at least one citation URL. Engines without native citations sit near zero by construction.
Grounded: Among answers that named the brand, the share whose mention was backed by a live citation (retrieval) rather than recalled from training.
Positive: Among answers that named the brand, the share the analyst classed as positive in tone.
Latency: Mean round-trip time of the engine's API for an audit prompt, in seconds.

Questions

About the data.

Is my brand's data in here?

Only as part of an average. The aggregate functions return counts and rates per engine; brand names, prompts and answers never leave the database through this page.

Why are some engines 'collecting'?

A rate over a handful of answers is noise dressed as a number. Each engine's rates appear once it has 25 answers in the window.

Why do citation rates differ so much between engines?

Because the engines differ. Perplexity returns citations natively; several others return prose without sources unless the answer happens to include a URL. The index reports what came back.

How often does it update?

The page is regenerated hourly from the live database.

Every audit you run makes the index better.

Run yours — the answers are yours, the aggregate is everyone's.

Start free audit See pricing

Free first audit
No credit card
Live engine data
Your data stays yours
Cancel anytime