Working with clients

Your client's feedback was written by ChatGPT. Now what?

More design feedback is arriving as AI slop: long, confident, and about a design the AI never saw. Why it happens, how to answer it, and how to get the real thing

The HopBack teamOctober 20266 minutes read

Key takeaways

  1. AI-written feedback is a workflow problem: the client had a lot to say and only a blank email to say it in.
  2. Reply to the real notes and ask for pointing, not paragraphs. Never ask whether they wrote it.
  3. Review on the thing itself and the blank email, and the paste, go away.

What it looks like when it lands

You send the staging link on Monday. On Wednesday the reply arrives, and it is twelve paragraphs long.

It opens with "Thank you for sharing the latest iteration." It talks about "visual hierarchy" and "cohesive brand storytelling" and "optimizing the user journey." It suggests the hero "could benefit from a stronger call to action" on a page where the call to action is a green button the size of a postcard. It recommends "ensuring mobile responsiveness" for a site you tested on nine devices. And it closes with "Overall, this is a strong foundation to build on."

None of it is wrong, exactly. None of it is about your design, either. It reads like feedback on a website in general, because that is what it is. There is a name for this now: AI slop. Plenty of words, very little in them.

You get on a call to go through it. The client cannot tell you which heading they meant, because they did not write the sentence about the heading. By the end of the call you have three real notes and a lost afternoon.

Why it happens, and why it is not a character flaw

Nobody does this to be difficult. Look at the position the client is in.

They have been asked to review a design, which is a skill they were never taught. They have opinions, but they are not sure which ones count or how to phrase them. They have an email open, which is a format that rewards length and formality. And they have a tool one tab over that turns "I think the top bit is a bit much and the menu is confusing" into six hundred words of structured feedback with headings.

From where they sit, the AI reply looks like the considerate option. It is longer, it is polite, and it sounds like something a professional would send. The client is trying to be helpful, and the tool has told them that helpful means thorough.

So the honest diagnosis is this: the client has a lot of feedback and nowhere good to put it. Email is the wrong container. Everything downstream follows from that.

How to tell real feedback from AI slop

You usually know within a sentence, but it helps to be able to say why, especially if a colleague thinks you are being unfair.

  • It is about websites, not this website. Generated feedback talks in categories: hierarchy, consistency, whitespace, accessibility. Real feedback talks in nouns: the photo of the warehouse, the word "solutions," the button under the price.
  • It cannot be located. Ask "which one?" and there is no answer, because there was never a specific one.
  • It recommends things that are already there. Make sure the site is responsive. Consider adding a contact section. The AI did not open the page, so it cannot see what is on it.
  • The register is wrong. Your client writes two-line emails with no capitals. This one has an executive summary.
  • It is confident about everything. Real clients hedge. "I might be wrong, but." Generated feedback never is.

One of these on its own means nothing. Three together and you can be fairly sure.

How to reply without making it awkward

Do not call it out. The client was trying to help, and "did you write this?" turns a design review into an argument about honesty. Two moves work better.

Ask for the pointing, not the paragraph. Reply with something like: "Thanks for this. To make sure we change the right things, could you tell me the three spots on the page that bother you most? Even just 'the top photo' or 'the second heading' is perfect." You have reframed the task from writing to pointing, which is the task they can actually do.

Pull the real notes out and confirm them. Somewhere in the twelve paragraphs are the two or three things the client actually feels. They are usually the specific ones. Quote them back: "I think the main things are the hero photo and the menu order. Have I got that right?" Nine times out of ten the answer is "yes, exactly," with relief.

What you are doing in both cases is taking the writing off their plate. They never wanted to write an essay. They wanted to be heard.

How to stop it happening next round

The reason the pasted reply exists is that the blank email exists. Take the email away and the problem mostly goes with it.

That means reviewing on the thing itself. When a client can open the actual design, click the heading they mean and type a sentence, there is nothing to write up. The click does the locating. The sentence does the rest. A note that would have been a paragraph in an email becomes "this heading is too long for mobile" with a pin on the heading.

A few other habits help:

  • Ask for a number. "Anything you want changed, pin it. Five pins is a great round." Permission to be brief is permission to be honest.
  • Give them a deadline, not an open window. Open-ended reviews invite the big write-up. "By Friday" invites the quick pass.
  • Send the reviewer guide with the link. One page on how to give useful feedback does more than any amount of coaching on the call. Ours is at For clients.
  • Say thank you for the short ones. "Five pins, all clear, all done by Thursday" is the behavior you want more of. Tell them.

How HopBack handles this

HopBack was built by an agency that got this exact email. The whole product is an answer to it.

A review in HopBack is the live site, or the PDF, or the video, with a comment box that opens where the client clicks. To leave a comment they have to look at the real thing and click the real element. The comment carries the element, the page and the screen size, so "the heading" is never ambiguous. There is no blank email, no document to fill, and nothing to paste a prompt into.

The result is feedback in the client's own words, because the format only accepts their words. Short, located, theirs. That is the feedback you can build a round from.

Questions people ask

Should I tell the client I think the feedback was AI-written?

No. Reply to the parts you can act on and ask them to point at the rest. The goal is better feedback next round, not a confession.

What if the generated feedback contains a good point?

Use it. A good note is a good note. The problem is not that AI wrote it, the problem is that most of it was not about your design. Confirm the real points with the client and move on.

Does a feedback tool actually stop AI slop?

It removes the container that causes it. When reviewing means clicking on the design, the client is looking at the design, and the path of least resistance is a short comment on the thing they clicked.

My client sends voice notes instead. Is that the same problem?

Same cause, different symptom. The client has feedback and no good place to put it. The fix is the same: give them the design and a way to point.

Send your next link and get feedback in your client's own words

14 days of Studio. No card needed.

One email a month, no more

Short, practical, about client feedback and running an agency. Unsubscribe in one click.