Perplexity is winning followers, but not the attention that should come with them
Not much. Perplexity’s posts together collect 15.54 likes and comments for every thousand followers. Anthropic’s collect 89.88 on the same measure over the same weeks. Of the 8 companies we can measure this way, Perplexity is 7th, and only Google AI is lower.
what the audience gives back in likes and comments
01The answer first, then how we measured it
Nothing on this page came from inside Perplexity. It is public information. We count followers every day on every platform that still gives us a number. We hold every post with the likes and comments it earned, the ads from three public ad libraries, and the YouTube videos that talk about each product.
The audience is growing. Perplexity added 330,588 followers from 6 April to 7 September 2026, twenty-two weeks, on the platforms where we can count followers, and holds 2,707,030 followers on those platforms. On a few other platforms we have a count but cannot watch it change week by week. Add those in and the total is 2,719,686, which is the figure in the table further down.
What it gives back is small. Perplexity’s posts together collect 15.54 likes and comments for every thousand followers. Anthropic’s collect 89.88 for the same thousand. Of the 8 companies we can measure this way, Perplexity is 7th, and only Google AI is lower.
The rest of this page is the three things that sit behind it: where the audience is, what the advertising asks for, and who is showing other people how to use the product.
02The audience is growing. Only one other company gets less back from each follower.
This counts the likes and comments on everything a company published in the eight weeks of posts this page measures, divided by the followers it had at the start on the platforms it posted to. It is the plainest way to ask whether an audience is paying attention.
A company needs at least eight posts in those weeks, and followers we can count week by week, before it appears here. 8 of the 10 companies clear that. The two that are left out do have posts, and they do have followers. What they do not have is a follower count we can follow across these particular weeks, and the note under the chart says which two they are.
Likes plus comments on everything the company published in the eight weeks from 6 July 2026 to 30 August 2026, divided by the audience it held at the start on the platforms it posted to. Mistral AI is left out because we have no follower count for it on the day this period starts, so there is nothing to measure it against. Microsoft Copilot is left out because the platforms it posted on in these eight weeks are not ones we can count followers on. The growth figure for it in the table below comes from other platforms and a longer period. Where we have no follower count for a platform, neither its posts nor its followers are counted here, so the post count on this chart can be lower than the post count in the table below.
8 companies qualify: eight or more posts, and a follower count we can follow across these weeks on the platforms they posted to.
Swipe the chart sideways to see all of it →
03Every company on the board, side by side
The follower column is the most recent count we hold. The two columns after it cover only the platforms where we have a count at the start date and at the end date, so the followers a company is shown gaining can come from fewer platforms than the followers it holds.
A percentage in this table is never a share of the follower count beside it. xAI is the clearest case: its growth figure is growth on the one platform we can follow, not growth on the whole follower count in that row.
| company | followers, most recent count | gained on the platforms we can measure | growth on those platforms | posts in 8 weeks | likes and comments |
|---|---|---|---|---|---|
| OpenAI | 27,463,853 | 4,021,044 | 17.3% | 392 | 1,077,400 |
| Anthropic | 7,546,366 | 3,078,217 | 69.8% | 102 | 393,525 |
| Google AI | 4,663,855 | 61,872 | 2.1% | 51 | 68,230 |
| Perplexity | 2,719,686 | 330,588 | 13.9% | 91 | 36,886 |
| xAI | 2,303,898 | 85,956 | 41.3% | 15 | 12,498 |
| DeepSeek | 1,314,013 | 147,920 | 12.7% | 9 | 86,907 |
| Mistral AI | 935,590 | not measurable | not measurable | 31 | 21,642 |
| Microsoft Copilot | 913,319 | 9,311 | 1.5% | 30 | 77 |
| Cohere | 420,750 | 59,958 | 16.7% | 146 | 28,478 |
| Glean | 261,734 | 48,292 | 22.7% | 185 | 12,234 |
The follower column is the most recent count on every platform we can count. The growth columns cover only the platforms whose follower history passes our checks, so the two columns cover different platforms and one is not a percentage of the other. Mistral AI has no follower history we can read across this window, so no growth figure. Some counts are older than others because a platform stopped giving us numbers: Perplexity (TikTok last counted 1 August 2026), xAI (X last counted 6 July 2026), Cohere (Facebook last counted 27 March 2026).
04Where Perplexity posts, and where the likes and comments come back
A follower is not worth the same everywhere, and most of this audience is on LinkedIn, where a follower gives back far less than one on X. Perplexity holds 1,520,680 followers on LinkedIn and 465,465 on X, counted on 6 July 2026, where the eight weeks of posts begin.
Each thousand of those LinkedIn followers produce 7.86 likes and comments. Each thousand on X produce 52.70. X gets 6.7 times as many back from the same number of followers.
LinkedIn takes 26.4% of Perplexity’s posts and gives back 32.4% of the likes and comments, so the posting there is doing its share. What is weak is the audience, not the effort.
Both bars are shares of the same eight weeks. LinkedIn returns a slightly larger share of the response than it takes of the posting, so the problem is not that Perplexity posts too little there. It is that most of its followers are on LinkedIn, and a LinkedIn follower leaves far fewer likes and comments than one on X. YouTube and Facebook get less back still, and they carry 6.6% and 1.1% of the posting.
91 posts by Perplexity, 6 July 2026 to 30 August 2026.
Swipe the chart sideways to see all of it →
05The obvious objection, answered inside one platform
You might think the gap is only there because these companies post in different places. Here is the same count taken one platform at a time. Posting in different places does not explain it away: the gap is still there inside a single platform.
Some of the gap is the mix, and not all of it. On X alone, Perplexity gets 52.70 for every thousand followers and Anthropic gets 170.63. The gap is smaller than it is overall, and it is still large.
Each company against its own audience on that one platform. On X, Perplexity is 6th of 7. On LinkedIn it beats OpenAI, because every company with a large LinkedIn following gets few likes and comments there. Only companies that posted on X in these weeks are here at all. Among those, a LinkedIn bar is drawn where the company also published on LinkedIn, and Mistral AI and Microsoft Copilot are absent from both, because on the platforms where they posted we hold no follower counts across these weeks. xAI is not on this chart because it did not post on X in these weeks, and this chart is built on the companies that did. The one figure xAI has, on LinkedIn, is 60.02, and it appears on the first chart on this page.
Every company that published on X in the eight weeks and could be measured there, 6 July 2026 to 30 August 2026.
Swipe the chart sideways to see all of it →
06The ads ask for an account before they make a case
Of the 1,038 Perplexity ads that show a button, 83.1% ask you to sign up or install something. For Anthropic it is 42.5% of 240 ads with a button. Only one company on the chart asks more often than Perplexity does.
Neither is right or wrong on its own. It is a choice about who the advertising is for: an ad that asks for an account is aimed at somebody ready to open one, and an ad that asks you to learn more is aimed at somebody who is not ready yet.
Out of the ads that show a button at all. The Google ad library publishes no button for search ads and LinkedIn publishes only the words printed on the button, never where it leads, so the share is worked out from the ads whose button we can see, never against every ad. The other ads with a button ask the reader to go and read something.
Counted on the ads that show a button: 1,038 of Perplexity’s 2,289, 760 of OpenAI’s 1,943.
Swipe the chart sideways to see all of it →
07Perplexity’s ads come down quickly
The typical Perplexity ad is taken down after 9 days. The typical Anthropic ad runs 270 days, which is past the right-hand edge of the chart below: at 180 days, 64 in a hundred of its ads were still running.
Perplexity has 1,950 ads that started early enough to have run ninety days. 4 in a hundred of them ran that long. For Anthropic it is 67 in a hundred.
An advertiser who keeps an ad running for months is usually telling you it works. One that replaces its ads every week or two is either testing constantly or not finding the one worth keeping, and from outside, those two explanations look the same.
Read left to right: of the ads that were old enough to have run this long, what share actually did. Asking the question only of ads old enough to answer it is the point: an ad that started five days ago cannot have run ninety, and counting it as a failure would invent the finding. Each point therefore rests on a different number of ads. A company whose ads are nearly all less than a month old is left off this chart. We would have too few older ads to say anything about how long its ads last.
Meta and Google only, where run dates are published.
Swipe the chart sideways to see all of it →
08What language the advertising is written in
We can read the words on 1,434 Perplexity ads. Of those, 23.3% are not in English. The languages say which countries a company is buying attention in, which is a different question from how much it is spending.
These shares are minimums, on purpose. The ad libraries never tell us what language an ad is in, so we work it out from the words themselves. We only call an ad non-English when it uses two words that belong to one other language and to no other, or an alphabet English does not use, which means it misses ads with very few words and almost never mistakes English for something else. The real figures can only be higher, never lower.
Out of the ads carrying at least fifteen letters of writing we can read.
Swipe the chart sideways to see all of it →
09Most of what Perplexity posts on X is replies
63.3% of Perplexity’s posts on X are replies to somebody else, and those replies collect 7.2% of the views Perplexity’s X posts get. Most of the posting effort on X goes where almost nobody sees it.
Almost every company here does this, so it is normal for the industry rather than a mistake only one of them makes. It is still worth knowing before treating a post count as a measure of effort.
Almost every company here spends most of its X output on replies, and every one of them collects a small fraction of its views from them. Replying is customer service. A reply is shown to far fewer people than a post of your own.
7 companies had enough X posts to score, out of 10.
Swipe the chart sideways to see all of it →
10On the platform where people go to learn a tool, far fewer are teaching this one
From 11 June to 9 September 2026, thirteen weeks, 219 YouTube channels mentioned Perplexity in a video, across 540 videos. Claude was mentioned by 1,767 channels.
This is the one number here that nobody can buy directly. A channel makes a video about a tool because its viewers want it, so the count is a reading of how much demand there is to learn the product.
Each channel is counted once for each product its videos name. We count the product and not the company, so this is ChatGPT rather than OpenAI, and Claude rather than Anthropic.
Coverage collected from 11 June 2026 to 9 September 2026.
Swipe the chart sideways to see all of it →
11The videos about Perplexity are also watched far less than the ones about the leading products
The 540 videos mentioning Perplexity collected 3,042,287 views between them. The chart below counts in millions, so that is the bar reading 3.0.
The same videos as the chart above, counted by how many people watched them. How many channels talk about a product and how many people watch them are two different things. The order barely changes between these two charts, but the distance does: the leaders pull much further ahead once you count who watched.
Videos with a usable view count. We left none of them out.
Swipe the chart sideways to see all of it →
12Three things worth trying
Grow the audience on X rather than moving posts there. A follower on X leaves 6.7 times as many likes and comments as one on LinkedIn, and Perplexity holds 465,465 on X against 1,520,680 on LinkedIn.
Leave a few ads up for longer. The typical ad comes down after 9 days, so almost none of these ads has been given the chance to be the one that works. This is the weakest of the three suggestions, because we cannot see why an ad stopped.
Give people something worth making a video about. 219 channels made a video mentioning Perplexity, against 1,767 for Claude. That gap is the clearest sign on this page that the product is not yet something people explain to each other.
Questions
Ask about anything on this page. Questions and answers are public. Every answer comes from the numbers on this page, and if those numbers cannot answer your question, you will be told so.
Every number on this page comes from tracKeting’s AI board: 10 AI companies. Anything you ask about this page is answered from the numbers on it, and where they cannot answer, you will be told.
the full study, with every chart and every calculation
Want to ask questions about this? This insight's id is
tracKeting tracks what founders, AI brands and Geneva estate agencies publish, and what it earns them. See what a board costs.