#e4mXplains:  When everyone optimises for engagement, who is actually engaging?

If even Mary Shelley's opening passage from Frankenstein is relegated to the AI-generative scrap heap, what hope do LinkedIn professionals have when it comes to their writing skills? Turns out, it may not even matter.

Bereft of watching the hour hands of the clock tick by, as most of us are these days, one of the most prominent metrics of productivity today is our LinkedIn engagement. The feed it gives us, the likes we get, the conversations we're supposedly having. But, as with so many other things in life today, AI is All Inside it.

Like the Mary Shelley meme doing the rounds, in which the alleged AI percentage varies almost as much as the number of times it is shared, there's a distinct sameness on the world's professional bulletin board. And it's symptomatic of a larger, increasingly AI-generated digital world.

As exchange4media explored previously, the cost of producing content has collapsed, while the human capacity to consume it remains stubbornly finite.

Read more: When everyone can make everything

AI can make infinite content but can it make anyone care?

An August 2026 NP Digital study of 100 companies found that fully AI-generated material accounted for 52.1% of new content but attracted just 4.9% of organic traffic. Producing more, evidently, doesn't guarantee anyone will read it.

But there's another complication. When everyone is optimising content for the same algorithms, does the resulting sameness eventually undermine the engagement those algorithms were designed to maximise?

LinkedIn's increasingly familiar feed

On October 20, Taranjeet Singh, Chief Commercial Officer at Ecosystm and former India head of Twitter, noted in a LinkedIn post that the platform was losing sight of relevance, recency and meaningful professional conversations.

Singh described recycled content, older posts resurfacing and fewer timely updates from people he actually follows. The professional network, he argued, increasingly resembles an "engagement optimisation engine rather than a professional intelligence platform."

His central question was whether optimising for engagement might actually be diminishing it.

When AI content gets a label, does it lose value? Read more here

LinkedIn, interestingly, has been grappling with precisely this problem.

In March 2026, it announced more advanced generative recommenders and large language models to improve feed relevance, personalisation and timeliness. By June, the company was acknowledging the proliferation of low-effort AI content, reporting that its early systems correctly identified generic material 94% of the time in initial testing.

On July 30, LinkedIn CEO Daniel Shapero acknowledged complaints about posts lacking insight and personality, with members saying the content "feels like AI slop."

LinkedIn subsequently introduced a "Seems like AI slop" reporting option, allowing members to flag generic or inauthentic posts and help train its detection systems. Because if there's one thing we love more than complaining about AI-generated content, it's thinking ourselves the best at figuring out what IS AI-generated content.

97 per cent for example.

By September 15, the company reported that views of content classified as AI slop had fallen 40%. More than one million members had used the reporting feature within its first two weeks. LinkedIn also said it was blocking hundreds of thousands of automated comment attempts daily.

Considerable technological effort, in other words, was being directed towards making the internet sound a little less like the technology increasingly responsible for writing it.

The numbers tell another story

On October 22, AI-detection company Originality.ai published research suggesting AI-generated writing remained widespread.

Analysing 109 randomly selected English-language LinkedIn posts of at least 100 words published in September, the company classified 85, or 78%, as likely AI-generated, up from approximately 76% in August.

LinkedIn registered the highest proportion among the platforms studied, followed by X and Threads at 62% each and Facebook at 54%.

Earlier research from Pangram, published in July, had classified approximately 41% of long-form LinkedIn posts exceeding 250 words as fully AI-generated. Its broader dataset covered more than one million social-media items collected through an opt-in browser extension.

There are caveats. Both firms sell AI-detection technology, their samples and definitions differ, and their classifications cannot conclusively establish authorship. Originality.ai's threshold included material judged likely to contain at least 15% AI-written text.

Nor do the studies necessarily contradict LinkedIn's reported improvements. The platform measures exposure to content it classifies as low-value; the researchers estimate AI authorship. These are not the same thing.

Which brings us back to Mary Shelley. Detecting machine-like writing is not necessarily the same as identifying whether a machine wrote it, much less whether it is worth reading.

The timing is particularly interesting for LinkedIn's parent, Microsoft, which is scheduled to announce its fiscal 2027 first-quarter earnings on October 28.

In its previous quarter, LinkedIn generated $5.19 billion in revenue, up 12% year-on-year, taking its FY2026 revenue to $19.8 billion. Its Marketing Solutions advertising business grew 16%, marking seven consecutive quarters of double-digit growth, while Premium subscriptions also maintained double-digit growth.

The engagement numbers were equally impressive. LinkedIn reported a 10% increase in time spent consuming content, 24% growth in knowledge-oriented posts and an 18% rise in time spent reading comments. Global membership recorded its fifth consecutive year of double-digit growth.

There is another interesting development. Beginning FY2027, Microsoft is reorganising its financial disclosures, folding LinkedIn's advertising and Premium businesses into its broader Search and Advertising reporting, while Talent and Sales Solutions move into Industry Solutions. LinkedIn will no longer have a standalone revenue-growth metric in Microsoft's standard quarterly investor disclosures.

The business is evidently growing, and its users are spending more time consuming content. Whether those increasingly measurable interactions amount to more meaningful professional engagement is an altogether different question.

Those are healthy commercial indicators. But they don't necessarily tell us whether the conversations taking place on the platform are becoming more meaningful. As Microsoft prepares to update investors on LinkedIn's performance, the distinction between engagement as a business metric and engagement as a user experience seems increasingly worth examining.

An internet optimising itself

The consequences extend considerably beyond LinkedIn.

On October 8, Tim Jurka, Vice President of Engineering at LinkedIn, offered an unusually candid account of the technological arms race unfolding within the platform.

As AI-generated content evolves, LinkedIn is using larger AI models to recognise emerging patterns of generic content, training smaller models to identify them at scale. Human reviewers intervene when the systems encounter particularly ambiguous material.

According to Jurka, the platform's classifiers now cover all posts distributed beyond users' immediate networks, with 94% precision in identifying what LinkedIn considers AI slop. That figure represents the company's reported detection precision, not the proportion of unwanted content successfully eliminated.

There's a certain irony to the arrangement. AI is being deployed to identify and suppress generic content produced with AI, while creators are increasingly using the same technology to make their content more appealing to the algorithms deciding what deserves attention.

An increasingly sophisticated technological arms race, with human attention as the prize.

Indian marketing leaders have expressed similar concerns.

And therein lies the paradox. Creators use AI to generate content tailored for platforms. Platforms use AI to determine what deserves distribution. Advertisers use automated systems to optimise their messaging against the resulting attention.

Those are healthy commercial indicators. But they don't necessarily tell us whether the conversations taking place on the platform are becoming more meaningful. As Microsoft prepares to update investors on LinkedIn's performance, the distinction between engagement as a business metric and engagement as a user experience seems increasingly worth examining.

The commercial cost of sameness

Even LinkedIn's advertising business is confronting the distinction.

The platform's AI-slop feedback option extends to feed advertisements, allowing users to flag creative they consider generic, repetitive or lacking substance.

However, LinkedIn says advertisers are not currently notified when their ads receive such feedback, nor is it available as a campaign-reporting metric. The information is being used internally to improve content quality.

For marketers, that distinction matters. A campaign may generate impressions, clicks and conversions without necessarily producing memorable communication or strengthening a brand's relationship with its audience.

None of this makes AI-assisted marketing inherently ineffective. Nor does human authorship guarantee originality. The question is whether the relentless pursuit of measurable efficiency is beginning to crowd out the less easily quantified qualities that make communication worthwhile.

LinkedIn's predicament is simply a particularly visible example of a larger systems problem.

AI makes more content possible. More content creates greater competition for attention. That competition encourages increasingly aggressive optimisation, prompting platforms to refine their algorithms further.

Everybody is optimising, everybody is publishing, and everybody is competing for engagement. Whether anybody is actually engaging is becoming a rather more interesting question.

Mary Shelley, at least, created something people are still discussing more than two centuries later.

Whether today's algorithms would recommend it is another matter entirely.