AI Chat is a conversational interface, not a fixed-query tool. Two factors cause variation:
​Session context
AI Chat reads chat history within a session, so earlier questions influence how later ones are interpreted. Asking "What's trending in serums?" after a question about Korean brands will weight differently than asking it in a new session.
​Synthesis is fresh each time
Even identical queries are re-synthesized from the underlying data each time. The data is the same, but the phrasing of the answer — what gets emphasized, how it's structured — can shift based on small differences in your wording.
The underlying data doesn't change between queries (unless a scheduled refresh runs between them). What changes is how AI Chat interprets and presents that data.

Be explicit in every query. Specify channel, time range, category, and comparison frame. "Top sunscreen brands" is ambiguous; "Top 10 sunscreen brands on Olive Young by review volume in April 2026" is reproducible.
​Start fresh sessions for unrelated questions. Don't chain unrelated queries in the same session — context bleeds between them.
​Use the same wording when comparing. If you ran a query last week and want to repeat it, paste the exact same prompt.
​Verify the data scope each answer used. Channels, time ranges, and category definitions can shift even with similar queries — confirming these match is more important than matching word-for-word phrasing.

For workflows that demand identical outputs run after run, AI Chat isn't the right tool.
​Generator is the alternative — it uses templated reports with fixed parameters. Run the same template every week, get the same structure with refreshed data. Because the query structure is locked, outputs stay directly comparable across runs.

I asked the same question yesterday and got a different answer — did the data change?
The data may have refreshed between queries, or session context and phrasing variation may be the cause. Compare the data scope shown in each answer to confirm. If scope matches and data hasn't refreshed, the variation is from synthesis.
​Why does adding one word change the answer so much?
AI Chat reads intent from wording. Adding "rising" vs. "top" vs. "trending" can shift which signals it weights. For locked outputs, use Generator with a fixed template.
​Can I save a "favorite" query?
This is being considered for future updates. For now, repeating the exact same prompt in a fresh session is the closest workaround.
​Does this happen in Generator too?
Generator outputs are more consistent because the query structure is fixed (you select analysis type, categories, and parameters rather than writing free-form prompts). The data refreshes between runs; the structure stays the same.