How to review your LinkedIn analytics with an AI assistant
With MediaPilot connected to an AI assistant, you can ask about your LinkedIn results in plain sentences and get answers drawn from your real figures. That is useful for exactly as long as you know what the assistant is reading: a summary for a period and totals for each post — impressions, reactions, comments, shares, clicks, engagement rate — and nothing else. Not who saw a post, not why it travelled, not anyone else's numbers. This guide lists the questions worth asking, in order, and what each answer can and cannot tell you.
Start with a period, and ask for two
“How did my LinkedIn posts do over the last 30 days, and over the last 60?” One period alone gives you totals with nothing to hold them against. The tool looks back from today — any number of days up to a year — and cannot start in the middle, so to see the month before last, have the assistant subtract the 30-day figures from the 60-day ones. For each period you get impressions, reactions, comments, shares and clicks, and an engagement rate: reactions, comments and shares as a share of impressions. That tells you whether the account as a whole is moving. It does not tell you why, and a higher total can be a single post.
Ask which posts carried the period
“Which posts had the most impressions?” The summary names the top posts by impressions, with the opening of each so that you can recognise them. Two cautions. Impressions rank reach, not usefulness; a post can be seen widely and move nobody. And a post's figures are running totals since it went out, so a post from three weeks ago has had three weeks to collect them and Tuesday's has had days. Compare posts of a similar age, or wait.
Then look at one post at a time
“Show me the numbers for the post about the pricing change.” For a single post the assistant returns its totals as of the latest daily update: impressions, reactions, comments, shares, clicks, engagement rate and click-through rate. It can fetch the full text as well, which is what makes the later questions possible. LinkedIn reports once a day, so a post published today has nothing yet and yesterday's has one day behind it — too early for a verdict on either.
Ask for a table, not an impression
The most useful request is the plainest: “List my published posts from the last two months with date, first line, impressions, comments and engagement rate, sorted by engagement rate.” The assistant has to fetch each post's figures in turn, so it takes a moment, but you end up looking at all of them side by side instead of at the few a summary picked. Then ask for the same table sorted by impressions. The two orders often differ, and the posts near the top of both are the ones to learn from.
Let it look for patterns, and treat them as guesses
Now the question a dashboard cannot answer: “Read the five posts with the highest engagement rate and the five with the lowest. What do the openings of each group have in common?” Because the assistant can read the text as well as the figures, it will find something — length, a number in the first line, a story against a list. Treat that as a hypothesis. Ten posts cannot separate the opening from the topic, the day of the week or luck, and a language model will readily produce a tidy explanation when asked for one. Use the answer to choose one thing to try on purpose in your next four posts.
Know what the assistant cannot see
Some questions sound answerable and are not, because the figures are not there. Who saw a post: job titles, companies, countries. How many followers you gained. What time of day people read. How your results compare with anyone else's. And posts written outside MediaPilot that were imported only for their analytics do not appear in the list of posts the assistant can browse. If an answer includes any of this anyway, the assistant is guessing; ask which figures the claim rests on.
End with one decision for next week
A review that ends in “interesting” has not finished. Close it with one sentence you can act on — “two posts that open with a client's question, both on Tuesday” — and say it in the same chat, where the assistant can read your topic pool and draft against the decision straight away. Then ask the same opening question in four weeks. The comparison between two reviews is worth more than either one alone.