LinkedIn is turning its own users into a quality signal for the professional feed, with more than one million people reportedly using its “Seems like AI slop” feedback option as generative tools make polished but generic workplace commentary extremely cheap to produce.

The reporting option was introduced in late July and allows users to identify material they regard as low-quality or inauthentic AI content. LinkedIn can use that response alongside other ranking signals rather than treating a single complaint as proof that artificial intelligence generated a post.

Chief product officer Hari Srinivasan has acknowledged publicly that AI-generated sameness has become a problem for the network, pointing to automated posting, fake profiles, and large quantities of text written in a common synthetic style. The company has been adjusting both ranking systems and its own generative features while giving member feedback a greater role.

Fresh reporting says more than one million people have now used the option and that posts receiving strong negative signals are seeing materially lower distribution. LinkedIn is also experimenting with giving creators feedback when audiences perceive their material as inauthentic, bringing the response out of an entirely invisible moderation process.

AI changes the economics of posting

The problem is unusually sharp on a professional network because regular posting can increase visibility among colleagues, customers, recruiters, and prospective employers. Before generative AI, producing a continuous stream of plausible professional commentary required somebody to write it or pay for the work; language models have reduced that marginal production cost almost to zero.

Automation therefore creates an incentive problem rather than merely an authorship problem. A useful post can be written with AI assistance, while entirely human-written material can still be repetitive, promotional, or empty. LinkedIn cannot preserve quality simply by detecting whether a model touched the text because provenance does not reliably measure whether the contribution contains useful experience or judgement.

The platform’s approach increasingly reflects that ambiguity. Srinivasan has described action against tools that automate posting for distribution growth and against fake profiles that use AI to appear authentic, while also recognising that legitimate users may use AI to improve grammar, structure, or accessibility.

That distinction has pushed LinkedIn away from encouraging wholesale AI rewriting and towards lighter assistance around proofreading and communication. Community feedback adds another signal between automatic detection and a blanket restriction on AI-supported writing.

Professional feeds have a trust problem

The quality issue extends beyond irritation because LinkedIn sits between social network, recruitment marketplace, sales channel, professional identity system, and publishing platform. If activity becomes heavily automated, signals such as posting frequency, comments, and apparent engagement become less useful as evidence that a real professional has contributed knowledge or built a genuine network.

Automated comments illustrate the difficulty. They can make a post appear more relevant than its underlying audience response warrants, while fake or heavily automated profiles can distort recruitment and prospecting. In that environment, verification and reputation become more valuable because ordinary content activity carries less information about whether an account is credible.

LinkedIn has been expanding member verification alongside its anti-automation work, creating another layer of evidence around identity even when individual pieces of text cannot be classified confidently. That approach recognises that the more important question is often not whether AI produced a sentence but whether the behaviour around the account is authentic.

The company is also confronting a contradiction created by the technology industry’s enthusiasm for generative content. Platforms spent several years embedding AI writing tools and encouraging users to produce more material, only to discover that removing production costs can flood recommendation systems with competent prose that carries little original information.

Detection gives way to behavioural signals

A sustainable response will probably depend on combining provenance, automation patterns, account reputation, audience feedback, engagement quality, and verification rather than trying to answer the narrower question of whether AI wrote a particular paragraph. That approach is less tidy than a binary label but better aligned with the actual quality problem.

Community moderation has risks of its own because users can mistake stylistic cues for evidence of AI and penalise polished writers, non-native English speakers, or people using assistive technology. The value of the feedback button therefore depends on LinkedIn treating reports as one signal among many rather than a direct classification mechanism.

The experiment offers an early indication of how enterprise-facing platforms may adapt as generated communication becomes ordinary. Organisations are already using models for recruitment messages, sales outreach, marketing posts, and executive commentary, while recipients are developing lower tolerance for text that feels automated even when it is factually unobjectionable.

LinkedIn helped normalise AI-assisted posting and is now trying to determine where assistance turns into noise. More than one million reports suggest users are willing to participate in drawing that boundary, although the harder challenge is ensuring that the ranking system rewards substance rather than merely teaching humans and models how to imitate whichever style is least likely to be flagged.