In late July, LinkedIn quietly added an option to the three dot menu on every post and comment: "Seems like AI slop." It was widely treated as a joke at the platform's own expense, given how enthusiastically the same platform had spent two years encouraging people to let a model rewrite their thoughts. Then the usage numbers arrived. The button has been tapped more than a million times, LinkedIn says members are now seeing around 40 percent fewer views on content its classifiers rate as slop, and the company has started notifying posters when their own writing gets flagged.
That is not a novelty feature finding an audience. That is a large professional network discovering that its users will happily do unpaid moderation work in exchange for a quieter feed. The complaint underneath it has nothing to do with whether a machine was involved. It is about padding: content engineered to look substantial, three hundred words where forty would do, confident phrasing wrapped around no actual information. People are not rejecting AI. They are rejecting noise, and they are voting for whoever removes it.
Aphelion AI is a private enterprise AI platform built to deploy AI agents inside infrastructure you own and control, keeping your data governed, your costs predictable and your tools shaped around how your business actually works. We built a feature on exactly the instinct LinkedIn just measured. It is called Clear View, and it does for your inbox what that button does for the feed.
What LinkedIn Actually Measured
Strip away the commentary and the result is a straightforward piece of user research conducted at scale. Given a one tap way to say "this is not worth my attention," over a million people took it within weeks. Nobody was paid, nobody was prompted, and the flagging happened on a platform where most users are there reluctantly in the first place.
The supporting changes tell you as much as the button does. LinkedIn retired its "enhance your post" tool, the one that rewrote your words into corporate cadence, and replaced it with a proofreader that leaves your voice intact. It has ramped up classifiers for low quality content, blocked large volumes of automated posting, and added safeguards so the flag cannot easily be turned into a weapon against people you disagree with.
The design lesson for anyone building business software is worth stating plainly. The winning move was not generating more content faster. It was removing content the reader did not want, and returning the human voice to the content that remained. Those are two different products, and most enterprise AI has been sold as the first one.
A million taps is a preference stated out loud. Readers will spend their own effort to reduce what reaches them, but almost none to produce more of it. Any AI feature that adds to the pile is fighting that current. Any feature that thins the pile is riding it.
The Same Problem Is Sitting In Your Inbox
Your feed is optional. Your inbox is not. The average knowledge worker opens dozens of business emails a day, and a startling share of each one is material nobody intended you to read. Consider what actually arrives with a typical supplier or client message:
- Signature furniture: logos, headshots, award badges, social icons, a mobile number you already have, and a two hundred word confidentiality notice with no legal force.
- Campaign residue: banner graphics, tracking pixels, unsubscribe links, event promotions and a satisfaction survey attached to a message about an invoice.
- Thread archaeology: the quoted history of eleven previous replies, each one carrying its own copy of all of the above.
- Padding in the body itself: two paragraphs of pleasantry and context setting wrapped around a single question that needs an answer today.
The cost is not dramatic in any single instance, which is exactly why it never gets fixed. It is three to five seconds of scanning per message to locate the actual point, repeated eighty times a day, plus the low grade attention tax of scrolling past the same corporate wallpaper again and again. Nobody has ever put that on a business case, and everybody feels it by four in the afternoon.
Clear View Is That Button, For Email
Clear View strips a message back to what is being said. The footer graphics, the signature block, the legal boilerplate, the tracking links, the survey request and the buried quoted history all come out. What is left is the sender, the point, and whatever is genuinely being asked of you. If there is an attachment or a date or a figure that matters, it stays. If it exists to fill space or to track your click, it goes.
The comparison with LinkedIn's button is close but not exact, and the difference is in your favour. On LinkedIn you flag noise after reading it, and the benefit lands on some future feed. Clear View applies before you read, on every message, so the saving is immediate and it happens whether or not you remember to do anything.
"A million people tapped a button to make their feed quieter. That is not an anti-AI backlash, it is a demand for signal. The most valuable thing AI can do inside a business is not write more, it is remove everything that was never worth reading and hand the rest back in half the time."
Stuart Smith, CEO, Aphelion AI
Then It Writes The Reply, In Your Voice
Removing noise solves the reading problem. The larger cost is on the other side, in the fifteen minutes spent constructing a reply that is accurate, correctly pitched, and consistent with how the business handles that kind of request. That is where most of the day quietly disappears.
Aphelion presents a short set of drafted replies for each message, and the point is what sits behind them. Each option is written in your own tone rather than generic professional cadence. Each one follows your documented company procedures, so the reply reflects how your business actually deals with that scenario. Each one carries the technical knowledge the answer requires, and each one states the correct next steps rather than a vague promise to follow up. You review the options, choose the one that fits, adjust a line if you want to, and send. An email that used to take ten or fifteen minutes to get right is finished in seconds.
That accuracy comes from grounding rather than guesswork. Aphelion's data enrichment capability turns your policy documents, procedure notes, product specifications and past correspondence into knowledge the platform can draw on, which is the difference between a fluent draft and a correct one. A generic assistant knows how business emails usually sound. A private deployment trained on your material knows what your business actually promises, what your warranty terms are, and what happens next.
| Handling one business email | With Aphelion | Standard inbox |
|---|---|---|
| Finding the actual point | Presented immediately | Scroll past footers and quoted history |
| Signatures, banners, surveys, legal blocks | Removed | Read around every time |
| Drafting a reply | Choose from prepared options | Written from scratch |
| Getting the tone right | Your own voice, learned | Rewritten two or three times |
| Following company procedure | Built into the draft | Look it up, or guess |
| Stating the next steps | Included and specific | Often vague or omitted |
| Typical time per considered reply | Seconds to review and send | Ten to fifteen minutes |
| Where the message content is processed | Your own infrastructure | Third party servers |
Figures for reply time are indicative and depend on the complexity of the message. The structural point holds regardless: the work of locating, deciding and phrasing is where the minutes go, and all three are addressable.
Why This Only Works On A Private Platform
Reading email properly means reading everything in it. Client names, pricing, contract terms, HR matters, legal advice, commercially sensitive negotiation. Handing that corpus to a hosted assistant means your correspondence leaves your control, sits in someone else's retention policy, and becomes a transfer you have to justify to a regulator and to your own clients.
Aphelion runs inside infrastructure you own or exclusively control, so message content never leaves your governed environment. Your obligations under GDPR do not disappear because you self-host, you remain the controller and the assessment is still yours to make, but it becomes a review of systems you can actually inspect rather than a vetting exercise on a black box. That is a meaningfully cheaper and more defensible position, and for firms in regulated sectors it is often the only version of this capability that clears approval at all.
There is a commercial argument alongside the compliance one. Hosted assistants meter you by usage, which means a feature designed to make your team communicate more efficiently produces a bill that rises every time it works. Aphelion is a flat per user fee, so the value of every minute saved stays with you rather than being clawed back by the invoice.
The Systems Angle Nobody Costs In
A drafted reply is only as good as what it knows. "I will look into that and come back to you" is a sentence any model can produce. "Your replacement unit shipped on Tuesday under case 4471 and the engineer visit is confirmed for Thursday morning" requires access to the systems where that is recorded.
This is why system integration is treated as a core Aphelion capability rather than a bespoke add on. Connected to the CRM, ERP, ticketing system and document stores you already run, the platform can ground a reply in the real state of the account, the order or the job. Without those connections you get confident text with no basis in fact, which is precisely the material people are now flagging on LinkedIn. The integration is what separates a helpful reply from a well written guess.
We are not selling a faster way to produce more text. Clear View removes what was never worth reading, and the reply options give you back a considered, procedurally correct answer in your own voice. Both run privately, on your infrastructure, on a flat per user cost, so the time you save stays saved.
Frequently Asked Questions
What is AI slop and why are platforms adding buttons to flag it?
AI slop is content that has been generated or padded out to look substantial while carrying very little actual information. It reads fluently, fills space, and leaves the reader no better informed than before. Platforms are adding flagging controls because the volume has reached the point where readers cannot find the useful material without help, and because the people doing the reading are the only reliable judges of what feels hollow. The underlying complaint is not about AI itself. It is about the ratio of signal to noise, and about who is expected to do the work of separating the two.
How does Aphelion AI Clear View remove the noise from business email?
Clear View strips an email down to the message it actually contains. Footer graphics, legal boilerplate, signature blocks, tracking links, survey requests, banner images and the long quoted history of a thread are all removed, leaving the sender, the point, and anything genuinely being asked of you. It is the same idea as a flag on a low-signal social post, applied at the moment you open the message rather than after you have read it. For anyone working through a hundred emails a day, the saving is not one dramatic hour, it is a few seconds of scanning recovered on every single message.
Is it safe to let an AI read company email under GDPR?
It depends entirely on where the reading happens. Email is one of the most sensitive data stores a business owns, holding client details, pricing, contracts, HR matters and legally privileged material, so routing it through a shared external service creates a transfer that has to be justified, documented and audited. Aphelion runs privately on infrastructure you own or exclusively control, which means the message content never leaves your governed environment and is never exposed to a third party's retention or training pipelines. Your GDPR obligations as a controller still apply, but the assessment becomes a review of your own systems rather than a vetting exercise on infrastructure you cannot inspect.
Can AI suggested replies follow our company procedures and connect to our systems?
That is the difference between a generic writing assistant and a private deployment. Aphelion draws on your own documented procedures, technical knowledge and policy material, so a suggested reply reflects how your business actually handles that situation, including the correct next steps rather than a vague promise to follow up. Because Aphelion connects to the CRM, ERP, ticketing and document systems you already run, a reply can reference the real state of the account or the job rather than inventing something plausible. Procedures and system connections are what turn a fluent draft into a reply you are willing to send.
Aphelion Clear View versus built-in AI email assistants: what is the difference?
Built-in assistants from the large platform vendors are hosted, metered and trained to sound like a generic professional, which is precisely the voice that people are now flagging as slop. They also require your correspondence to leave your environment to be processed. Aphelion runs privately on your own infrastructure, learns your tone rather than replacing it, applies your procedures rather than generic best practice, and carries a flat per-user cost instead of a bill that rises with every message you process. The practical test is simple: one produces text that sounds like everyone else, the other produces a reply that sounds like you and stands up to your own compliance review. You can read more about the team behind the platform on our About page.
Signal Is The Product
The most useful thing about LinkedIn's experiment is that it settles an argument. For two years the assumed direction of travel was more content, produced faster, by more people. A million taps in a few weeks says the market for that is thinner than anyone thought, and that the genuinely valuable position is on the other side of the exchange, with the reader.
Inside a business, the reader is your team, and the feed is the inbox they cannot ignore. Clear View removes what was never worth their attention. The drafted replies hand back the minutes that used to go into phrasing something correctly. Both run privately, grounded in your own procedures and systems, on a cost that does not rise the more useful it becomes. That is what applying AI to a real business looks like: less noise, faster answers, and a voice that still sounds like yours.