Playbooks · 5 min read
Audit Your Posts by Questions Asked, Not Likes Earned
We do not have an analytics product. No dashboards of reach, no engagement scoring, no audit tool, no report that grades your last ninety days. Nothing in this product reads your posts at all — it reads comments, and only to decide whether to reply to them.
So this is not a post about a feature. It is about a way of auditing an account that we happen to have a strong opinion on, because it is the audit that predicts whether any of this will work for you. Almost every content audit sorts posts by likes, saves or reach. The more useful sort, for anyone considering automation, is by how many comments asked a question.
Why the usual sort misleads
Reach tells you the algorithm liked something. Saves tell you somebody intended to come back. Neither tells you whether a human being was moved to speak.
A post can perform beautifully on every metric and generate no conversations. A post can flop on reach and fill your inbox. Those two posts should be treated completely differently, and a reach-sorted audit puts them in the wrong order.
High reach, no questions
Modest reach, many questions
The audit exists to tell those apart, and it takes about an hour by hand.
The audit, done by hand
You do not need a tool for this and we are not selling one. Open your notifications and your last thirty posts.
- Read the comments as text, not as engagement. Fifty at a time. You are not counting them, you are classifying them.
- Sort every comment into four buckets. Question — anything asking for information you would have to answer. Compliment — nice, and nothing follows from it. Tag — somebody tagging a friend. Noise — emoji, bots, tag-spam.
- Write the question bucket down verbatim. Not paraphrased. The actual strings:
price?,kitna hai,dm me,available in Pune?,size chart bhejo. You are building a vocabulary list, and paraphrasing destroys the thing that makes it useful. - Score each post by its question count alone. Ignore the other three buckets for ranking purposes. One number per post.
- Look at the top five posts and find the common property. It is usually one of three things: the post showed a specific product, the post withheld one piece of information, or the post explicitly asked people to comment.
That last step is the finding. Everything before it is data collection.
What the verbatim list is for
The list of exact strings you wrote down in step three is the most valuable artefact of the audit, and it has two uses.
The first is trigger construction. People type the shortest thing that will plausibly work, and the words they choose are not the words you would have guessed. Nobody writes "what is the price of this item". They write price?, cost kya h, kitna, or just ?. Building an any-of list from strings you have actually received beats building one from imagination every time, and the keyword post goes into the matching rules.
The second is deciding what to automate at all. Group the verbatim list by intent and count the groups. If four or five intents account for most of what you receive, automation will cover a large part of your inbox with templated replies, which cost nothing. If your questions are all different from each other, it will not, and you should know that before you spend an evening building triggers.
The three buckets you are not automating
Worth being explicit, because people try.
Compliments need a human or nothing. An automated thank you so much! under a genuine compliment is the single most detectable form of automation there is, and it costs you more goodwill than the reply earns.
Tags are somebody doing your marketing for you. Leave them alone. There is nothing to say to a person who typed a friend's handle.
Noise should be actively excluded rather than ignored. If tag-spam under your giveaway posts contains a word that is in one of your trigger lists, add it to negative keywords so it does not consume a serious reply.
Auditing the answers, not just the questions
The second half of the audit is the part people skip, and it is where you learn whether the answers are stable.
Take your top three question intents and write out, from memory, the answer you would give. Then go find three real replies you sent to each and compare.
If your written answer and your real replies match, that intent is templatable — free and unlimited, no credit spent, and it will read exactly like you because you wrote it. If they diverge, look at why. Usually the answer depends on something specific to the asker, and that intent belongs to a person, not a template.
Do it again in three months
The reason to repeat it is that the question distribution shifts when your content shifts, and trigger lists rot quietly.
You launch a new product and price starts meaning a different price. You start shipping to a new city and delivery questions change shape. Nothing warns you; the automation goes on confidently answering last quarter's question.
An hour every quarter, reading your own comments as strings, is the whole maintenance burden. If the audit tells you a large share of your inbox is repeated questions with stable answers, the setup guide is the next thing to read. If it tells you otherwise, you have saved yourself the effort, which is also a good outcome.
