The Creative Insights agent analyses your ad performance at the level of individual creative elements. It finds the patterns behind your results, shows you what is winning and what is failing, and turns those patterns into specific recommendations for your next round of creative work.
You chat with the agent in plain language. Every finding it reports traces back to a metric in your data, together with the tag, time window and sample size behind it.
Where do I find it?
Select Agents in the left-hand menu and open the Creative Insights agent from the library.
What data does it need?
The agent analyses only the data you give it. It works from either of two sources.
A performance file you upload, such as a CSV or spreadsheet with creative-level metrics (click-through rate, conversions, spend, impressions) and tag columns.
The ad accounts you have connected to Pencil. You can read more in Why should I connect my ad accounts to Pencil?.
If neither is available, the agent asks you to upload a file or connect an ad account before it analyses anything. It never invents numbers to fill a gap.
What can I ask it?
Ask anything you would ask a creative performance analyst. For example:
"Which creative tags are working in last quarter's Meta performance?"
"Why is the new sustainability angle underperforming?"
"Compare hook styles across TikTok and YouTube for the last 30 days."
When you ask a broad question, the agent scans across the full set of tag dimensions rather than silently narrowing its scope. These dimensions include:
Format: video, static, carousel, story, UGC, motion graphic.
Theme: product-led, lifestyle, problem-solution, testimonial, social proof.
Hook: question, statistic, demo, before-after, pattern interrupt, creator.
Call to action: shop now, learn more, sign up, or none.
Audience: prospecting, retargeting, lookalike, broad.
Platform: Meta, TikTok, YouTube, programmatic and others.
Duration: for video, the length cohort, such as six seconds, 15 seconds or 30 seconds.
Visual style: colour palette, lighting, on-screen text density, talent type.
What does a response include?
Each analysis opens with its data context: the source, the platforms and creatives covered, the date range and the comparison point (your account average, a prior period or another tag). The agent then reports its headline findings, the tag-level patterns behind them, and your top and bottom performers.
It closes with three to five recommendations for your next creative cycle. Each one names the tag to double down on or retire, the metric it should move, and the size of the change the data implies. The recommendations are specific enough to hand to a creative team as a brief.
[Screenshot: a Creative Insights analysis in chat, showing headline findings and recommendations.]
How does it keep findings reliable?
The agent is built to resist overclaiming.
It states the sample size (impressions, spend or creative count) alongside every rate and lift figure.
It flags patterns drawn from small samples as directional rather than treating them as a verdict.
It surfaces contradictions instead of hiding them, for example a tag that wins on click-through rate but loses on conversion rate.
It treats lift as an association in your data, not proof of cause.
It does not use industry benchmarks or competitor data. It only analyses what you upload or connect.
How is it different from the Insights dashboard?
The Insights dashboard is where you browse tags, creatives and metrics yourself, using filters, sorting and search. The Creative Insights agent works conversationally on the same kind of performance data and goes a step further: it interprets the patterns and recommends what to make next. Use the dashboard to explore your data. Use the agent when you want answers and a brief.