I run three sites: johnarce.com, aipriceradar.com, and tipidnation.com. The first two are built around AI tools and AI infrastructure, so getting cited by an AI Overview feels almost on-brand. tipidnation.com is different. It's a Filipino deals and promo site, credit card offers, Shopee codes, bank promos, built for someone hunting a specific voucher before checkout, not someone asking an assistant to explain a concept.
That difference is exactly why I didn't assume the AI-referral setup I wrote up in my GA4 and Search Console AI traffic piece would behave the same way here. A promo-code query and a how-to query hit AI systems differently. I wanted to see where the overlap actually held before I rebuilt anything.
Why a deals site doesn't play by the same AI-citation rules
Most of the queries that land on johnarce.com are explanatory: "what is X," "how do I do Y." Those map cleanly onto the answer-first structure I covered in how I get AI Overviews to cite my pages. A promo query is sharper and more time-sensitive: "is this code still active," "which bank has the better cashback this month." The answer isn't a definition, it's a yes/no plus an expiry date, and it goes stale fast.
That changes what "citable" means on a deals page. An AI Overview or a ChatGPT answer pulling from a promo page needs to trust that the page was checked recently, not just that it's well-written. Freshness stopped being a nice-to-have on tipidnation.com and became the main lever.
The tracking setup I carried over, and what I added
I reused the same GA4 custom channel group from the traffic piece above: session source/medium matched
against chatgpt.com, perplexity.ai, gemini.google.com,
copilot.microsoft.com, labeled "AI Referral." No changes needed there, the referrer logic
doesn't care what kind of site it's attached to.
What I added was a secondary dimension: page type. On johnarce.com every page is roughly the same shape. On tipidnation.com I needed to split AI-referral sessions by landing page category, single-offer pages, category roundups, and comparison guides, because I suspected they wouldn't earn citations at the same rate. They didn't.
What changed on the pages themselves
I didn't touch the comparison-guide pages much, they already read like how-to content and behaved like the pages on this site. The single-offer and category pages got rebuilt around four changes.
| Element | Before | After |
|---|---|---|
| Opening line | Promo described in a sentence buried under a headline | Direct status first: active or expired, then the code, then the expiry date |
| Freshness signal | No visible "last checked" date | Visible "last verified" date plus matching dateModified in schema |
| Structured data | Generic Article schema only | Offer schema with validThrough, nested under the Article |
| Comparisons | Prose paragraphs listing competing promos | Table ranking promos by value, same pattern that works on this site |
The schema layer: Offer schema beats a generic Article tag
Every page on this site runs Article or BlogPosting schema, that's the right shape for a how-to or an
essay. A promo page isn't an article pretending to have an offer attached, it's an offer, so I switched
the markup to lead with Offer, nested under the Article rather than the other way around.
The fields that matter most are validThrough and availability. Without them,
an AI system parsing the page has no structured way to know the promo expired, it has to infer that from
prose, which is exactly the kind of guessing that gets a page skipped rather than cited. With them, the
expiry is a machine-readable fact sitting next to the human-readable "last verified" date, so the two
signals reinforce each other instead of one being the only source of truth.
I also kept the Person and Organization schema from johnarce.com consistent across all three sites,
same name, same sameAs links. AI systems resolve "who's behind this" before deciding how
much to trust a claim, and a deals site benefits from that entity consistency just as much as a
personal blog does, arguably more, since trust is the whole product on a page telling someone to hand
over a payment method.
AI-referral sessions on tipidnation.com went from essentially nothing to a small, steady slice of total traffic, growing week over week rather than spiking and vanishing. I'm not going to quote a multiplier here. The raw volume is still small enough that any precise percentage would look more exact than it is. What I can say with confidence is the shape of the curve: it's the same gradual climb I described on this site, not a flat line.
What's different about watching this at affiliate-site scale
johnarce.com has a few dozen pages. tipidnation.com has hundreds, and merchant offers churn constantly, a code that's live today can expire next week with zero notice from the merchant. Rebuilding every page by hand the way I rebuild a blog post here isn't realistic.
So I triaged instead of rewriting everything at once. High-traffic, evergreen categories (recurring bank promos, ongoing cashback programs) got the full rebuild first: answer-first structure, Offer schema, visible freshness date. Seasonal and sale-event pages got a template applied instead of a hand edit, same structural elements, generated from a shared layout. One-off, low-traffic single codes stayed as they were. They weren't worth the time against the likely payoff.
A blog post's facts don't change after I publish. A promo code's status changes on the merchant's schedule, not mine. The freshness date is only honest if something actually re-checks the offer, so the harder problem isn't writing the answer-first opening once, it's keeping hundreds of them true on an ongoing basis.
Key takeaways
- The AI-referral tracking setup (GA4 channel group, GSC Generative AI report) ports to a deals site unchanged, the referrer logic doesn't care about content type.
- What citable content looks like changes by query type: a promo page needs a trustworthy freshness signal more than it needs elegant prose.
- Comparison tables outperformed prose on tipidnation.com the same way they do on this site, that pattern seems to hold across site types.
- At affiliate-site scale, triage by traffic tier beats trying to rebuild every page by hand at once.
- Directional growth, not a precise multiplier, is the honest way to report this while the raw numbers are still small.
What I'll do next
I'm building the re-verification step into the deploy pipeline so freshness dates update automatically when a scraper confirms a code still works, instead of relying on me noticing a promo expired. That's the piece most likely to break trust if I skip it, a visible "last verified" date that silently stops being true is worse than not showing one at all.
I'm also watching whether the comparison-guide pages, which I left untouched this round, start showing up in the AI-referral channel once the single-offer pages around them are fully rebuilt and linking into them. If internal linking into a cluster matters as much on an affiliate site as it did on this one, that's a cheaper fix than rebuilding every guide from scratch, and I'll know within a few weeks of watching the GA4 channel and the GSC Generative AI report together.
FAQ
Does the AI-referral GA4 setup work the same way on an affiliate or deals site? +
Why does freshness matter more on a promo page than a blog post? +
Did comparison tables help on a deals site the same way they help on a blog?
How do you decide which promo pages to rebuild first on a site with hundreds of pages?
Can you share the exact AI-referral traffic numbers for tipidnation.com?
Want this built for your team?
I design AI agents and growth automation that run without babysitting. If that sounds useful, let's talk.
Get in touch โ