ChatGPT ads: can we actually track them yet?

tagmanager

I've been seeing ChatGPT ads crop up more frequently, but tracking their performance is a proper nightmare compared to typical paid channels. A colleague tried their first run and got a flood of demo requests-yet analytics slapped most of it as direct traffic. No clear source, no attribution data. It's like throwing money into a black hole.

Even if we could measure results, that's not the same as having native tracking to optimise bids. For my clients, I won't touch ad spend without basic tracking in place. Some might do it just for the board bragging rights, but that feels hollow when you can't prove ROI. Until they sort out UTMs or at least some form of attribution, I'm out.

olivia-roas

If clients want to burn budget on "being on ChatGPT" just so they can show the board a screenshot, that's their funeral. Impressions are about as useful as a chocolate teapot when you're trying to prove actual business impact.

The real problem? They'll never get proper search term reports because every query is a unique snowflake - contextual and long-tail as hell. Bundling into semantic themes sounds nice until you realise that's just audience targeting with extra steps. Performance marketers are going to spend months fighting for basic attribution while the platform figures out what "conversion" even means.

Honestly, I'd only touch it for upper-funnel experiments with a strict cap. Anything else is just feeding the vanity metric machine.

neonlights

The real ROI killer here is attribution blindness-if you're not measuring it, you're losing the argument for budget. In most cases, users convert later via branded search or direct traffic, so ChatGPT gets zero credit. Runs through Google Analytics? Just lands in direct, unless you've built custom tracking or dedicated landing pages. That's a massive gap in the data. Without UTM parameters or a proper measurement framework, you're essentially flying blind on one of the fastest-growing channels.

analyticsjunkie

honestly, sounds like your mate just botched the setup. You can absolutely sling UTMs on those paid ad URLs - or drop their pixel/CAPI in and measure inside their own attribution world. It's not rocket science, just basic tracking.

jamesaso

UTM tracking in AI chat is still a black box for attribution. Until that gets sorted, dedicated landing pages are the only way to isolate performance - especially for app install campaigns where every click cost matters. Without them, you're just guessing which chat thread actually drove the conversion

buggedout

No surprise here. ChatGPT is clearly gatekeeping their traffic data and likely stripping UTMs and other tracking parameters until they've got enough internal data to prove performance.

Even if you could get coarse conversion reports, that's not the same as native tracking that lets you optimise bids and audiences.

For my clients, no chance I'm putting budgets behind ads without at least basic tracking in place. A few points:

Without UTM stripping they make attribution modelling impossible.
No visibility means you can't tell if it's cannibalising existing organic traffic or genuinely incremental.
Optimisation becomes blind - no way to shift spend toward what actually converts.

I'd need to see some real control group data before I'd recommend a test.

prpro

Attribution is going to be a bit of a headache with this one, at least for now. the way i see it, treat it less like a standard paid channel and more like dark social or influencer traffic. That means clean UTMs wherever possible, dedicated landing pages, promo codes or offer variants, and a proper before/after look at branded search and direct conversions.

The real challenge is proving incrementality. a spike in demo requests looks great, but without a holdout group or some kind of geo or time split, it's far too easy to give all the credit to the shiny new channel

brandvoice

feels a lot like the early days of dark social, doesn't it? you can see the ripple effects-branded search bump, direct traffic spikes, more demo requests-but clean, reliable attribution is still a proper mess. Makes it hard to prove what's actually driving results

marketingmule

Between the lack of proper pixel tracking and the absolute headache of that favicon "give us the image hosted on your domain" nonsense for logos and 1200x1200 sizes, I just can't get on board with the $50 USD CPM they're charging here in Australia. That's insanely expensive for reaching general consumers, not B2B decision-makers.

Any other Aussies here actually considering or already running GPT ads?

sleepermode

I've been digging into the StackAdapt / OpenAI integration myself. That's likely the most accessible on-ramp for teams without a six-figure test budget-otherwise the barrier is genuine. Beyond that, I'd categorise ChatGPT ads as another consideration-stage vehicle. The real challenge is attribution: without a first-party data pipe or deterministic tracking, you're flying blind on whether those impressions actually move the needle down funnel.

digitalnomad

This is bringing back flashbacks to the early dark social days - someone sees something, then converts later from "direct" traffic and you're left scratching your head. 👀

Honestly, I don't think most attribution systems are built for conversational discovery yet. People bounce between devices, or just remember the brand name and type it in later. 😅

My bet is most teams will lean on:

Lift tests 🧪
Blended metrics like brand search uplift or assisted conversions

Clean click attribution? Forget it for this channel. We're all still guessing, if we're being honest. Super early days - feels like everyone's making it up as they go.

ranktracker

Attribution for ChatGPT traffic is a nightmare to track properly, and I've spent more hours than I'd like pulling log files to try and make sense of it. The core issue is that users treat it more like a recommendation engine than a traditional ad click-through - they'll see your brand mentioned in a response, then go off and do their own research. That means you get a surge in direct traffic, cross-device sessions, and branded searches that don't tie back to any UTM parameter.

What I've found works best isn't trying to force last-click attribution into a model that doesn't support it. Instead, I track overall lift in branded search impressions from Google Search Console, plus any uptick in direct traffic alongside demo requests or inbound form fills. I also set up a custom segment in my analytics to flag any session where the referrer is something like chat.openai.com or similar - even if it's sporadic, you can start seeing patterns over a few weeks.

If you're a log file nerd like me, pull your server logs and look for user-agent patterns that might indicate bot vs real human traffic from ChatGPT's infrastructure - it's messy, but it gives you a sanity check. Honestly, the best KPI right now is "are we seeing more people searching for us and then converting without a clear source?" If yes, it's probably working. Just don't expect perfect attribution any time soon.

maya-b2b

Honestly, I think ChatGPT attribution is going to feel a lot more like podcast advertising or word-of-mouth than traditional Google Ads. Someone picks up a recommendation, wanders off, then comes back later via search or direct traffic-and analytics just shrugs and labels it "direct." If you're waiting for clean last-click attribution like Meta or Google, we're not there yet. AI-powered discovery is messy right now, and trying to force it into a last-click box is a recipe for frustration. In my pipeline work, I've noticed that people are more likely to research before reaching out after an AI mention, so I track assisted conversions and branded search lift instead. That's been a better indicator.

24fire

Careful not to treat that demo spike as evidence of ChatGPT ad performance unless the links were tagged meticulously. OpenAI's own documentation makes it clear: advertisers only get aggregated metrics like views and clicks-no chat-level or user-level data. attribution there is naturally thinner than what you're used to from Google or Meta.

For SaaS, I'd approach it like an emerging intent channel rather than a mature paid one. Dedicated landing page, strict UTMs, a CRM source field, hidden form fields, self-reported attribution, demo quality scoring, pipeline value, and close rate. If it shows up as direct traffic, that doesn't mean the channel failed-it means the tracking path is leaky.

Across several B2B client accounts, platform-reported clicks matter less to me than what happens once a lead lands in HubSpot or Salesforce: booked demo rate, SQL rate, ACV, sales notes, pipeline created. ChatGPT traffic could be high intent, but without clean tagging and CRM matching, you're essentially staring at direct traffic and dressing it up with nicer words

weekendhustle

honestly, it's still such a moving target for ChatGPT ads, even for someone like me who spends half their life testing new channels 😅.

for a national home services client, I'd suggest starting with a modest daily budget-somewhere between £50-£100 per day-if you're going regionally first. That gives you enough data to spot a rough CPA within a week or two, without burning through cash if the targeting's off. if you're brave enough to go after 6-10 high-performing states from the get-go, I'd probably double that daily spend just to get meaningful volume across markets.

as for benchmarks, honestly, i haven't seen anything solid shared publicly yet. From the little I've run, the CPC seems to land in the £0.50-£1.50 range, but conversion rates are all over the place depending on the offer. It's still very much "test and learn" territory. would love to hear if anyone else has cracked a better starting point?

davidsearch

Any word on India rollout? Need to confirm if the attribution gaps are identical across markets - can't justify scaling into a new geo without solid CPA data.

emmareach

Attribution is something they've clearly been working on. I signed up for the self-serve option to test it myself, and I know a few other brands who were in the initial beta. They're now saying conversion tracking and CAPI are live.

I'll believe it when I see the data, but there's no reason it shouldn't work - assuming they've actually built the plumbing properly. Wouldn't touch it for outbound without solid event-level tracking, though.

marcuthread

Oh mate, you've hit the exact nightmare. The moment that traffic goes through an AI search engine, tracking parameters just vanish into thin air. Standard analytics shoves it all into Direct, which is basically useless. Honest advice? The only thing that's worked for me is slapping a simple "How'd you find us?" dropdown on the demo request form. It's not fancy, but at least you get real answers instead of guessing.

harper-local

Not really. No proper attribution tools for it yet-just guesswork.

chloeest

Installed the pixel and split-tested it. the data showed a significant conversion lift - worth testing for yourself.