What is AI Citation Intelligence?

AI Citation Intelligence measures whether AI answer engines cite a brand, which source domains support the answer, and which competitors appear for buyer questions.

How is Findabl different from SEO tools?

SEO tools model rankings and backlinks. Findabl checks AI answer engines directly and records citations, source domains, and competitor mentions.

What does Findabl track?

Findabl tracks ChatGPT, Gemini, Claude, and Perplexity responses at the prompt level, including brand citations, competitor mentions, source domains, and weekly changes.

Is the AI Recommendation Stability report gated?

No. The AI Recommendation Stability white paper is available as an ungated download.

AI Recommendation Stability Study

Why this study matters

Buyers increasingly ask AI assistants for recommendations instead of comparing a page of search results. This changes the visibility problem for brands. A single AI answer can name a short list of vendors, source domains, or service providers, and the buyer may never inspect the wider market.

The question behind this study was simple: once a brand earns an AI citation, does it stay cited? The answer was no. AI recommendation is not a fixed position. It behaves like a shifting probability that changes by prompt, engine, retrieval source, and time.

How the study was run

The study was pre-registered before data collection. Research questions, hypotheses, the query set, segment assignments, engine choices, experimental conditions, extraction rules, and primary metrics were locked in advance.

Engine	Model version	Query coverage
ChatGPT	GPT-5.5	26 queries
Gemini	Gemini 3.5 Flash	26 queries
Claude	Claude Sonnet 4.6	14 queries
Perplexity	Perplexity Sonar	26 queries

Condition	What it means	What it isolates
Floor	Minimum randomness, no live web search	Model-level instability
Sampled	Normal randomness, no live web search	Generation variation
Web	Normal randomness, live web retrieval on	Real-world retrieval variation

Finding 1: AI recommendations are not stable

Even with randomness minimized and live web search off, identical questions returned the same top brand only 78% to 87% of the time, depending on the engine. With live web retrieval on, consistency dropped to 48% to 68%.

Engine	RCS floor	RCS sampled	RCS web
ChatGPT	0.78	0.78	0.48
Gemini	0.78	0.67	0.54
Claude	0.87	0.69	0.50
Perplexity	N/A	N/A	0.68

Finding 2: Engines often disagree

Across 26 buyer questions tested on all four engines, only 7 questions, or 27%, produced unanimous agreement on the top brand. A brand can lead on one engine and be absent on another, which makes single-engine tracking incomplete.

Finding 3: Most brands appear rarely

Only 2.4% of brand-question combinations reached a 100% appearance rate. 85.7% fell below 50%, and the median appearance rate was 10%. In practice, a brand that "appears in AI" may show up in 1 of 10 responses, not 9 of 10.

Finding 4: Gatekeeper publications shape the answer

The study found 376 unique cited domains across all engines and questions. The top 10 domains accounted for 28.6% of all 3,059 source citations. Chambers, Forbes, and TechRadar alone accounted for 15.8%.

Publication	Times cited	Citation share	Who it matters for
Chambers	250	8.2%	Legal services
Forbes	122	4.0%	Finance, business, consumer
TechRadar	110	3.6%	B2B software, consumer tech
U.S. News Best Law Firms	70	2.3%	Legal services
Tom's Guide	67	2.2%	Consumer tech, software
RTINGS.com	67	2.2%	Consumer electronics
NerdWallet	51	1.7%	Personal finance
Legal 500	49	1.6%	Legal services
Vault	47	1.5%	Legal, career
Sleep Foundation	43	1.4%	Consumer health, mattresses

What marketers should do next

Measure appearance rate, not a yes-or-no snapshot.
Track across ChatGPT, Gemini, Claude, and Perplexity.
Run repeated checks because one-off scans miss instability.
Identify the source domains AI engines cite most in your category.
Build coverage in those gatekeeper publications, then monitor drift.

AI Recommendation Stability