Which Clicks Came From ChatGPT (and Which Are Bots)

Your link report says 400 clicks. You screenshot it, you feel good, you move on. But pause for a second: how many of those 400 were real people? How many were a messaging app quietly fetching a preview card? How many were ChatGPT or Perplexity opening your page to answer someone else's question? Most link shorteners cannot tell you, and the standard click report only deepens the illusion: they hand you one number and let you assume it means 400 humans. It almost never does.

And the mix is shifting under your feet: the automated share of link traffic keeps climbing as more preview fetchers, security scanners, and AI agents crawl the open web, so the gap between your raw number and your real one is getting wider, not smaller. A short link is just a redirect, and everything that touches that redirect gets counted by a naive tool: real browsers, automated bots, email security scanners, and now a fast-growing wave of AI agents. If you make decisions on the raw total, you are budgeting, bragging, and optimizing on a figure that is partly fiction. The fix is not to distrust analytics, it is to split that total into the visitors that actually make it up.

What a raw click count actually contains

Every link you share in public gets automated traffic within minutes, often before a single human sees it. Post a link in a group chat and the app fetches it to build a preview thumbnail. Send it in an email and the recipient's security gateway pre-opens it to check for malware. Publish it anywhere crawlable and Googlebot, Bingbot, and a dozen social crawlers come knocking. None of those are customers, yet a shortener that counts raw hits files every one of them under "clicks". On a busy link, bots can easily outnumber people in the first hour.

The three-way split most tools never show

Once you accept that a raw number is a blend, the useful move is to separate it into three buckets: real people, automated bots and scanners, and AI assistants. The first is your actual audience. The second is noise you want removed. The third is genuinely new: clicks that originate from an AI tool answering a question, which behave like a referral channel of their own. Flyn now surfaces all three instead of mashing them into a single count, and the rest of this guide is about reading each one.

Before you can act on the split, it helps to know exactly who sits behind each bucket. These are not abstract categories, they are specific, identifiable kinds of request that hit your short link every day.

Real people, the number that matters

A real person is someone who tapped your link in a browser or an app and was sent to your destination. This is the number every other metric should be measured against. When we talk about a verified human click, we mean a request whose signals look like a genuine visitor rather than a known automated agent. It is the figure worth putting in a report, because it is the closest honest proxy for "someone actually saw this".

Bots, scanners, and link previews

This bucket is broad and mostly harmless, but it is not your audience. It includes link-preview fetchers (the bot that builds the thumbnail card in iMessage, Slack, WhatsApp, or Telegram), email security scanners that pre-open every link in an inbox, and search and social crawlers indexing the web. A few are malicious, most are routine plumbing, but all of them inflate a raw count. Our guide on click fraud prevention digs into the nastier end of this spectrum.

AI assistants, the new arrival

The newest kind of visitor is an AI agent: ChatGPT, Perplexity, Claude, Gemini, or Copilot fetching your page so it can cite or summarize it in an answer. Sometimes that is the AI browsing on a person's behalf in real time, sometimes it is the assistant collecting a source. Either way it is not a normal human visit and not quite a junk bot, so Flyn breaks it out as its own channel rather than lumping it in with either. That distinction is the whole reason "which clicks came from ChatGPT" is now a question you can answer.

Here is where honesty matters more than marketing. It is tempting to say a filtered number is "100 percent human" or "guaranteed real". It is not, and any tool that claims otherwise is overselling. What Flyn actually does is more modest and more defensible.

What the filter looks at

On every request, Flyn inspects the request signals, mainly the User-Agent string plus related headers, and matches them against a maintained list of known bots, crawlers, preview scanners, and AI agents. A request that identifies itself as Googlebot, or as a known preview fetcher, or as an AI crawler, is classified accordingly and kept out of the human total. Everything that does not match a known automated signature is counted as a human click. This runs inside the redirect, in a few milliseconds, on every link.

The honest limit: this is not identity verification

User-agent screening removes automated traffic at scale, and it is remarkably effective because the vast majority of bots identify themselves plainly. But it is not identity verification of any single visitor. A determined bot can spoof a browser user agent, and a privacy tool can strip headers from a genuine person. So the right way to read the human number is "clicks that were not identifiably automated", not "provably individual humans". That is a strong, useful signal for cleaning up your analytics. It is not a lie detector, and we will not pretend it is.

Flyn shows the split as a Traffic Quality view on your link. At a glance you see what share of the traffic was human, how much was filtered as automated, and how much came from AI assistants, instead of a single count you have to take on faith.

The bot-excluded human total is free

The single most useful number, the human click total with bots removed, is free on every link, on every plan. You do not have to pay to stop counting robots. If all you want is a click count you can trust, our link click counter and the standard click analytics already give you the filtered figure rather than the inflated raw one.

The split is free too, the sources are Pro

The breakdown itself costs nothing. The exact human-versus-bot split, the bot-filtered analytics engine behind it, and the verified human clicks view are all on every plan, free included. What Pro adds is where that traffic came from: the AI Assistants channel and the geo, device, and referrer breakdowns. The rule of thumb: how many of your clicks were real is free, where the real ones came from is Pro.

The part everyone wants to see: clicks that came from an AI tool. When someone asks ChatGPT or Perplexity a question and the assistant cites a page you shared, the request to your link carries signals that identify it as AI traffic. Flyn groups those into a single AI Assistants channel, sitting right alongside Search, Social, Direct, and Email.

Why AI referrals show up now

Two years ago this channel would have been a rounding error. Now a real share of research, shopping, and how-to queries start inside an assistant instead of a search box, and when the assistant hands back a link, the click lands on you. Before Flyn broke this out, those hits mostly disappeared into Direct traffic (the "we have no idea where this came from" bucket), which is exactly where useful signal goes to die.

Concretely: someone asks an assistant for the best tool for a job, or how to do something, and it hands back your page as part of the answer. That tap is an AI referral. It used to arrive with no fingerprint, indistinguishable from a pasted URL or an old bookmark. Now it has a name, a count, and a trend line you can actually manage.

What an AI-assistant click means, and does not

An AI-assistant click tells you your content or product is being surfaced inside AI answers, which is worth knowing and increasingly worth optimizing for. What it does not tell you is that a human definitely followed through: sometimes the agent is fetching your page to read it, not because a person clicked. Read the channel as a demand signal ("AI is putting me in front of people"), not a guaranteed visit, and watch its trend the way you would any referrer. If a particular guide or product page pulls steady AI traffic, that is a strong hint about what assistants consider a good answer, and a nudge to make more of it.

This is not a cosmetic cleanup. A count that is a third robots does real damage to the decisions you make on top of it, and the damage compounds quietly because the number still looks plausible.

Inflated numbers waste budget

If you judge a paid campaign by raw clicks, bots quietly pad your winners and your losers alike, and you scale spend toward whatever attracted the most automated noise. Strip the bots and the real cost per human click can look very different, sometimes dramatically so. The same applies to affiliate and creator deals priced on clicks: our affiliate link tracking guide and the Instagram ad tracking walkthrough both lean on the filtered number for exactly this reason.

They hide your real winners

Bots do not spread evenly. A link that got shared into a few big group chats can rack up hundreds of preview fetches, making a mediocre post look like a hit while a genuinely strong one that happened to dodge the scanners looks weak. Compare on real human clicks and the leaderboard reorders, often surprisingly.

A raw click count is not wrong so much as unreadable. Until you separate people from bots from AI, you are averaging three different stories into one number and calling it a result.

The inflated number does not stay in your dashboard, either. It goes in the report to your boss, the case study on your site, the screenshot in your pitch. Every place the raw figure travels, it carries the same fiction. Fixing it at the source, on the link, is the only way to keep the honest number honest all the way downstream.

Seeing the split is step one. Acting on it is where the value is. Here is what to do with each part once you can finally see it.

If AI assistants are sending you clicks

Lean in. A steady AI Assistants channel means your pages are being cited in answers, so make those pages easy to quote: clear headings, direct answers near the top, and facts an assistant can lift cleanly. Point your best "start here" content at a single tracked short link so you can watch the AI channel grow over time. This is the new, quieter cousin of SEO, and it rewards the same clarity.

If bots dominate a link

First, do not panic: most bot traffic is harmless preview and crawler activity, and the human total already excludes it. But if one link is drowning in automated hits, check where you shared it (mass email and big public channels attract the most scanners) and lean on the filtered number rather than the raw one for every decision. If you suspect genuinely abusive traffic, our click fraud prevention guide covers the next steps.

Pair the split with per-link tracking

The split gets sharper when each channel has its own link. Give every campaign, placement, or partner a distinct short link and add UTM parameters, and you can read Traffic Quality per source instead of in aggregate. Then you know not just that a link was 65 percent human, but which post, email, or Instagram placement brought the real people. Read side by side, a source that is 80 percent human beats one that is 55 percent human even at the same raw total, and the per-source view is the only place that gap shows up.

One last piece of honesty, because it decides how to read these numbers. The three buckets here, human, bot, and AI, are click measurements: they tell you how many real people opened your link. What happened after the click is a separate question, and Flyn now answers that one too.

Clicks are the foundation, conversions are optional

If you need to tie a click to a purchase, a signup, or a dollar amount, enable conversion tracking (a Pro feature) on the link and add a small tag to your own site: each goal then reports back against the exact link that sent the visitor, counted against these same verified human clicks. Without it, Flyn stops at the click and does that part honestly. Either way, knowing which number you are looking at keeps your reports honest.

How to turn it on

There is nothing to configure. Shorten a link, share it, and open its stats: the human total is already bot-filtered, for free. Traffic Quality is not a separate product, it is a layer on every link you already make. A branded link on your own custom domain (a Pro feature), a cloaked affiliate link, a smart link with routing: each one gets the same human, bot, and AI split. Upgrade to Pro for the full breakdown, the AI Assistants channel, and the geo, device, and referrer detail, then create a free account and check the real number behind your next link. From there, tracking which posts drive sales takes the same honest click number and joins it to revenue.