AI commentary
Having the AI comment on a period's numbers or a single post, where the commentary is stored, the prompt setting, and why the panel may not appear.
Last updated: September 30, 2026
Having the AI comment on a period's numbers or a single post, where the commentary is stored, the prompt setting, and why the panel may not appear.
Last updated: September 30, 2026
Looking at numbers is one job; seeing what they are telling you is another. AI commentary reads the metrics for the period you selected, or the performance of a single post, and explains in two or three plain paragraphs what happened, why it may have happened, and what to try next. This guide explains how the commentary is produced, where it is stored, and how its instruction is changed.
⚠️ Note: Generating a commentary spends AI credits. That is why it is started manually and never runs just because you opened the screen. Generating one requires automation edit permission; a user who can only view reports will read an existing commentary but cannot produce a new one.
The panel offers AI commentary in two places, and they work differently.
It sits in the column to the right of the chart on Insights › Overview. The panel's description says: "The period's numbers are read by AI; the commentary is stored for this period."
Only numbers are sent to the model: the period boundaries, the totals and change percentages of the ten metrics, the follower count, the highest and lowest reach days, and the top five buckets of the audience breakdowns. No account name, contact detail or message content is sent.
It sits on a post's analysis page, reached from Insights › Posts. Its description differs: "The caption, metrics and cover image are read by the AI; the commentary is stored for this post."
Three things go to the model here: the post's details and caption, its metrics, and its cover image. The image is sent because the model has to see the content to answer "why did this post work". For video posts the cover frame is read, not the video itself.
The steps are the same in both places:
Once a commentary exists the button is renamed Comment again. Commenting again overwrites the old text; two commentaries are never kept side by side.
A byline sits under the text: which model wrote it and when.
This is the most commonly misunderstood point:
When no commentary exists yet, the panel reads: "No commentary for this period yet." That does not mean the data is empty.
A collapsible box called Commentary prompt setting sits under the commentary panel on a post's analysis page. It is only drawn for users who may generate commentaries.
Its rules are:
The default instruction asks for two short paragraphs about the post: what it is about and how the numbers say it was received, plus one concrete suggestion to try next time. It also tells the model not to invent information, not to comment on metrics it was not given, and not to treat a metric marked "unknown" as zero.
The period commentary's instruction cannot be changed; it is fixed.
At the end of September a boutique selects the Last month period and clicks Comment with AI. The commentary says three things: reach rose 18% while saves barely moved; most of the rise is concentrated in two days (from Meta's highest-reach-day figure); and its suggestion is to repeat the content format that worked on those two days throughout the month.
The same boutique then opens its widest-reaching post and generates the post commentary. Because this one also sees the caption and the image, it says "the absence of pricing in the caption may have driven up the comment count" — an inference the period commentary could never make.
The AI commentary panel is not drawn at all in four cases. You will not see an empty box either; the panel simply is not there:
If the panel is there but shows only an explanatory line instead of a button, a permission is missing: the line reads "Generating one requires automation edit permission."
Used in the right place, the commentary turns the time you spend reading numbers into time spent making decisions.
Was this article helpful?