Last updated · report format 2.1

How lookfluence gets its numbers

Every number in a lookfluence report comes from public Instagram data and a fixed method. This page explains what the numbers mean, how to read them and what we can't see.

Where the data comes from

A lookfluence report is built only from data that Instagram shows publicly: the profile and its latest posts.

Profile and posts

For each report we read the public profile and its latest feed posts, newest first. Collab posts are read too, since they appear in the creator's feed. From the profile we take followers, following, the number of posts, the bio, the link in bio, the account type and category and public contact details. From each post we take likes, comments, views, the caption, the image, tagged accounts, the tagged place, the paid partnership label, whether comments are turned off and the time it was posted.

When we collect it

Data is collected when a report is built or refreshed, and every report shows that date. Anyone can refresh a report once it is 7 days old. If a refresh fails, the previous report stays and the next try is possible after 24 hours. Every report is kept as a dated snapshot, so older versions stay available.

What we never collect

For a private account we read only the public profile card, so its report has no engagement numbers, flags or score. A report never reads a profile's list of followers or anyone's messages.

Engagement rate

Engagement rate (ER) measures how many likes and comments a profile's posts get for its size.

Which posts count

Four kinds of posts are left out:

  • Collab posts that another account owns. They show in the feed, but their likes and comments belong to that account.
  • Pinned posts, which are often old and keep collecting likes.
  • Posts with hidden likes, because their real engagement can't be seen.
  • Very fresh posts, which are still collecting likes. If too few older posts are left, the fresh ones stay in, so the rate isn't measured on one or two posts.

The newest of the remaining posts are measured. The report shows how many posts were measured and which were left out.

When most posts hide likes

If most of the profile's own posts hide likes, likes can't be measured. The posts are then measured for comments and views only, and the report has no engagement rate, no rating and no score. The red flag checks that need like counts are skipped.

The formula

Each post's engagement rate is its likes plus comments, divided by followers. The report leads with the typical post: the median of these rates, the middle one when they are lined up. The average of the rates is shown next to it.

Post engagement rate = (likes + comments) ÷ followers

Typical post = median of the post rates · Average = mean of the post rates

Example: five posts at 1.0%, 1.2%, 1.5%, 2.0% and 9.0%. The typical post gets 1.5%, the average is 2.9%.

Typical post and average

One viral post can lift the average a lot but can't move the typical post. When the average is far above the typical post, one or two posts did most of the work, and the report doesn't rate engagement.

Photos, videos and reels

Photos and carousels, feed videos and reels each get their own typical post and average by the same formula. All of them divide by the same follower count, so they can be compared with each other and with the overall rate.

Engagement per view

For videos and reels the report also shows likes and comments per view: the total likes and comments divided by the total views. Posts that show no views are skipped rather than counted as zero, and because the totals are added up first, a post with a handful of views can't swing the result.

Reels reach and viral share

Reels reach is the median views of the measured reels that show views, divided by followers. The same median is the typical views of a reel. Viral share is the share of those reels with more views than the profile has followers. Reel views are Instagram's play count.

Comparison with accounts the same size

Engagement depends on size, so we compare each profile only with accounts the same size. Two numbers are compared: the engagement rate of a typical post and reels reach.

Follower groups

Every profile falls into one of seven groups by followers. Each group includes its lower bound: a profile with exactly 10,000 followers is Micro.

Follower groups
GroupFollowers
Under 1KUnder 1,000
Nano1,000–9,999
Micro10,000–49,999
Mid-tier50,000–499,999
Macro500,000–999,999
Mega1,000,000–9,999,999
Giant10,000,000 and more

Percentile and median

The percentile is the share of profiles in the group with a lower value: 78 means 78% of the group has a lower engagement rate on a typical post. The group's median is its 50th percentile, with half of the group above it and half below. Reels reach gets its own percentile the same way.

Rating words

The report turns the percentile, rounded to a whole number, into a word. Reels reach uses the same words.

Rating words by percentile
PercentileWord
80 and aboveHigh
60–79Above average
40–59Average
20–39Below average
Under 20Low

Engagement gets no word when likes are hidden, when too few posts were measured, when one or two posts did most of the work, or when there is no comparison.

Where the groups come from

The groups are built from public profiles we have analyzed recently, one entry per profile, and are refreshed regularly. They describe the profiles that pass through lookfluence, not a random sample of Instagram. A profile counts only when enough of its posts could be measured and its likes are visible; for reels reach, only profiles whose reels show views count. A group needs enough profiles before we compare anyone with it. Without an engagement comparison the report has no rating word and no score.

Red flags for bought followers

Five checks look for signs that likes or followers were bought. Each flag is a reason to look closer, not a verdict: the checks are set so that a flag on a real account stays rare. The checks use the same posts as engagement rate.

Very few comments for the likes

Raised when posts get unusually few comments compared with their likes. Real audiences usually comment more; likes without comments are typical of bought likes. Not checked when likes are too low to judge or when many posts have comments turned off.

Unusually even likes

Raised when likes are almost the same on every post. Organic posts swing widely, purchased like packages come out flat.

Engagement above almost all similar accounts

Raised when engagement is higher than almost every account of the same size. This is either exceptional content or engagement that was bought or exchanged in engagement groups. It is informational and doesn't lower the score. Very small profiles and profiles without a group to compare with are not checked.

Follows more accounts than follow back

Raised when the profile follows far more accounts than a typical creator, and more accounts than follow it, which is typical of follow-for-follow growth. It is informational: it lowers the score less than a warning.

Reels reach below almost all similar accounts

Raised when reels reach is among the lowest in the group: a typical reel is watched by a much smaller share of followers than at accounts the same size. Very low reach suggests followers who are inactive or not real.

Warnings and informational flags

Very few comments, unusually even likes and low reels reach are warnings. Follow-for-follow and engagement above almost every account of the same size are informational. When most posts hide likes, the checks that need like counts are skipped.

Account Quality Score (AQS)

Account Quality Score (AQS) is a 0–100 score. It is built from four parts, each scored from 0 to 100. Engagement counts most. Red flags then lower the result: a warning more than an informational flag.

AQS parts
PartWhat it measures
Engagement vs similar accountsHow the engagement rate of a typical post compares with accounts the same size. It counts most.
Reels reach vs similar accountsHow far reels reach compared with accounts the same size.
Posting activityHow recently and how regularly the profile posts.
Profile completenessHow complete the profile is: bio, link, account type and contact details.

The score never goes below 0, and engagement above almost every account of the same size doesn't count against it. A part that can't be measured, such as reels reach for a profile without reels, is left out, and the other parts are scaled up to fill its place.

There is no score when likes are hidden on most posts, when too few posts were measured, or when engagement has no comparison: the other parts alone would rate how well the profile is kept, not its quality.

Texts written by a language model

The verdict at the top of a report, the paragraph about the profile, what it is a good fit for, what to watch out for, the notes on the best and weakest posts, the niche, topics, brands, the caption language and two answers in the questions section are written by a language model. It reads the captions and images of the latest posts and the numbers above, already calculated. It is told to use only these inputs, to state no number it wasn't given and to say nothing about who the followers are: we have no audience data.

A brand counts only when a post shows it: an Instagram paid partnership label, a tagged brand account, a brand named in a caption or a branded product clearly seen in an image. The profile's own products, shows and shops are listed as its own projects, not as brands.

How the texts are checked

Before a report is published, every text is checked against the numbers we calculated and against the posts it mentions. A text that fails is replaced with a fixed sentence built from the numbers, or left out. If the model fails altogether, the report is published without its texts. A private profile, or one without a post to measure, gets no texts. A language model can still be wrong about what a post shows, so check the niche and brands against the posts.

Paid partnerships

Paid partnerships are not guessed: a post counts as one when Instagram labels it a paid partnership or tags a brand partner on it. Co-authors of collab posts are listed separately. When enough measured posts carry the label and enough don't, the report compares the typical post of each group.

Growth and posting activity

Follower growth comes from our earlier checks of the same profile. The report shows the change over about a month and about three months when we have a check that old; without one that change is not shown. Over a very short period a change says more about noise than growth, so it isn't shown. A profile checked for the first time has no growth yet.

Posting activity is read from the feed without pinned posts. Posts a week is the number of gaps between posts, divided by the days they span, times 7, using the recent posts. The report also shows the date of the last post. Dates are in UTC, since we don't know the creator's time zone.

Example: 12 posts over 28 days make 11 gaps, and 11 ÷ 28 × 7 = 2.75 posts a week.

What we don't know

We can't see inside private accounts, so their reports stop at the public profile card. We don't see who follows a profile, so there is no audience age, gender or location. Places in a report are the places tagged in the posts: they show where the creator posts from, not where followers live. The caption language comes from the captions, not from the audience. We can't count fake followers either: the five flags above are signs, not a count.

Changes to this method

Report format 2.1, . Method updated: more posts are read, the typical post (the median) is now the main engagement number, comparison groups come from our own archive, and the score has a new formula. Reports built before 2.1 keep their numbers until the profile is refreshed.

Report format 2.0. Engagement is measured without pinned, fresh and hidden-like posts; profiles are compared with accounts the same size; red flags and the the Account Quality Score were added; niche, brands and the summary are written by a language model. Reports built before 2.0 keep their earlier numbers until the profile is refreshed, and their old score is not shown.

Who makes lookfluence

lookfluence is built and run by a small independent team. Reports are calculated automatically by the rules on this page; lookalike searches and deep analyses are run by hand for now. For a question about the method or a number that looks wrong, write to us through the contact page or at info@lookfluence.com.