How to Read Your Facebook and Instagram Insights With AI

By Leon Harris, Founder & AI Creative Director · Updated September 2026

You open Meta Business Suite once a month, see a wall of numbers under Insights, and close the tab without changing anything about what you post next. That's not a discipline problem — it's that raw reach and views numbers don't tell you what to do differently. The fix isn't checking insights more often. It's turning last month's numbers into one prompt an AI can actually reason about, and one change you commit to for the next 30 days.

Where the numbers live now

In Meta Business Suite (business.facebook.com or the app), open Insights in the left menu, then the Content section for per-post numbers, or the overview/results section for account-level trends. On Instagram, open your profile's professional dashboard — as of 2026 it's split into three tabs: Overview (views, reach, profile visits, follows), Engagement (interactions, including newer share and skip-rate numbers), and Audience (who's seeing you). Instagram only keeps about 90 days of history in-app, so export monthly if you want to compare further back.

A label changed — don't panic if your numbers jumped

Meta's developer blog announced in August 2025 that it was retiring the Impressions metric across the Page Insights API in favor of Views, to match what the Facebook app and Business Suite already show. Trade coverage puts the effective rollout for reporting tools around June 2026. The practical difference: Views count every time your content appears on screen, including repeat views by the same person, so a post can show a higher number this year than an equivalent post would have shown last year under the old Impressions count — without actually reaching more people. If your dashboard suddenly looks better (or a connected reporting tool breaks), this rename is the likely reason, not a real change in performance.

Exporting your post-level data

In Insights > Content, set your date range (last 28 days, or widen it), then use the Export Data button near the date filter to download a CSV of your posts and their metrics. This CSV, trimmed down, is what you'll hand to an AI chatbot.

The 3-4 numbers that matter — and the ones that don't

Not every number in the export deserves attention. For a small business, weight these higher: reach or views from non-followers (shows the post traveled beyond your existing audience), shares and saves (a viewer choosing to forward or keep something is a stronger signal than a passive scroll-past), messages started or link clicks (this is the number closest to an actual inquiry), and follows gained from a specific post (shows it worked as discovery, not just engagement). Likes and comments alone are the weakest signal here — they're cheap to give and don't require the viewer to do anything that touches your business.

Prepping the export for a chatbot

Open the CSV and cut it down to five columns: post date, format (photo, reel, carousel), the first line of the caption, and your 3-4 chosen metrics. A typical month's worth of posts for a small page is a tiny file, well inside any major free chatbot's upload limits — though those limits do shift (ChatGPT's free tier, for one, has capped uploads to a handful a day at various points in 2026), so if a file upload gets fussy, just paste the trimmed rows directly into the chat as text instead.

A prompt that won't just tell you what you want to hear

The prompt matters more than the tool. Paste your trimmed data and ask for this, specifically:

"Group these posts by format and topic. Compare each group against my own average for the period — not an industry benchmark. Name my top 3 and bottom 3 posts, and for each one, give your best reasoned guess at why it performed that way. Then tell me how confident you are in that reasoning, given how many posts I've given you."

That last line matters. Left alone, AI will confidently invent a cause for what's often just noise — ask it to state its confidence and it will hedge appropriately on a small sample.

Turn the answer into one change

Don't rewrite your whole content plan from one month of data. Pick the single most repeatable difference the AI flagged — a format, a topic, a caption style — and commit to it for the next 30 days. Then re-run the same export and prompt and compare. Not sooner: a week or two isn't enough posts to separate a real pattern from a lucky one, and Instagram's 90-day window means you shouldn't wait much longer either.

Why messages started should outrank likes

For a Philippine SMB that closes sales over Messenger, a post that generates a handful of started conversations did more for the business than one with ten times the likes and zero inquiries. Likes measure attention; messages started measure someone deciding to talk to you. If you're only tracking one number month to month, make it that one.

What this loop can't tell you

Be honest about the limits. A small page's numbers are noisy — one post that happens to catch a share chain can skew your whole monthly average, so don't over-read a single outlier. Platform metric definitions change, as the Views rename shows, so don't assume this year's numbers mean exactly what last year's did. And insights never see what happened after the conversation moved into DMs, or a sale that closed offline weeks later — they measure exposure and the start of interest, not revenue. For the content plan and ad-spend side of this, see our guides to building a 30-day AI content calendar and running Facebook and TikTok ads on a small budget.

Frequently asked questions

How often should I run this insights review?

Once a month is enough. Reviewing weekly usually just reacts to noise — one post can skew a small page's short-term numbers. Monthly gives the pattern time to show itself, and it lines up with a 30-day "change one thing and check" cycle.

Do I need a paid tool to export my Facebook and Instagram data?

No. Meta Business Suite's built-in Insights > Content section has an Export Data option that downloads a CSV for free — no third-party analytics subscription needed for this loop.

Why does my Views number look so much higher than my old Impressions number?

Meta replaced the Impressions metric with Views, and Views counts repeat exposures to the same person, while Impressions was counted differently. It's a change in what's being measured, not proof your reach suddenly grew.

Which matters more for a small business: reach or messages started?

Messages started, if you sell through Messenger. Reach just tells you how many people saw something; messages started tells you someone decided to act on it. A post with modest reach but several started conversations is doing more for the business than a high-reach post with none.

Can AI reliably tell me why a specific post did well or badly?

It can offer a reasoned guess, not a proven answer — with a small monthly sample, differences are often just noise. Always ask the AI to state its confidence given the sample size, and treat a single month's read as a hypothesis to test, not a verdict.

What if my free chatbot won't accept my CSV export?

Trim the file down to the columns you actually need (date, format, caption opener, chosen metrics) and paste the rows as plain text in the chat instead of uploading the file. Free-tier upload limits change and vary by tool, so this is the more reliable fallback.

Want this done for you? We build the AI content system; you approve. Free audit + custom quote within 24 hours.

Get a free audit →