ChatGPT, Perplexity and AI Search as Traffic Sources for Affiliate Sites: What Gets Cited, What the Referral Data Shows, and Whether to Change Anything
A growing share of product questions are being asked of an assistant rather than typed into Google, and the assistant answers with a few sentences and a short list of sources. For an affiliate site this is a new kind of result: not ten blue links, but two or three citations under a synthesised answer, with a click rate that depends on whether the reader wants to verify.
The traffic is small so far. The visitors are unusually ready to buy. And the question of what to do about it has a shorter answer than the industry around it suggests.
This post covers how assistants cite sources for product questions and how that differs from Google's AI Overviews, what the referral share looks like for product sites and who sends it, and the page features that get cited and the ones that don't.
It also covers how to see the traffic in analytics (including the part that hides as direct), and why the practical answer is to write the honest guide rather than to optimise for a new channel.
The AI Overviews post covers Google's version; this is the assistants outside Google's results page.
The short answer
AI assistants send a small, high-intent share of traffic to product sites, citing pages that make specific, attributable claims with a stated method, an author and current data. Make sure the site is crawlable by their bots, write verdicts that can be quoted, keep prices and picks current, and build an analytics segment for the known referrers.
Don't build separate "AI-optimised" content; the features that get cited are the features a good guide already has.
How assistants cite, against Google
| Google results (with or without an AI Overview) | AI assistants (ChatGPT search, Perplexity, Copilot, Gemini) | |
|---|---|---|
| What the user sees | A list of pages; an Overview sometimes above it with links | A written answer with two to eight numbered sources |
| How a page gets in | Ranks for the query | Is retrieved for the question and quotable for a claim in the answer |
| How many sources | Ten per page | A handful; being one of them is a larger share |
| Click rate | Varies by position; depressed by Overviews | Low overall; higher when the answer is a recommendation the reader wants to check |
| Who clicks | Everyone from browsers to buyers | Mostly people verifying or about to buy |
| Freshness | Index-based; days to weeks | Live retrieval for many answers; today's prices matter |
| What the engine quotes | Title and snippet | A specific sentence or claim, attributed |
The last row is the one that changes writing: an assistant doesn't cite a page for being about the topic, it cites a page for a sentence it can attribute. A guide whose claims are specific enough to be quoted gets quoted.
What the referral data shows
- Share: for most product sites, low single digits of Google's referrals, growing quarter on quarter. Perplexity has been the most consistent citer of independent sites; ChatGPT's search mode is the fastest-growing source; Copilot and Gemini are smaller.
- Landing pages: deep guide and comparison pages, almost never the homepage. The visitor arrived with a question already answered and wants the detail or the button.
- Intent: high. Engagement and click-through to Amazon on assistant referrals run above Google's organic average on the same pages in most reports from product sites, because the assistant did the browsing stage.
- Hidden traffic: some assistant visits arrive without a referrer and land in "direct". A rise in direct sessions landing on specific guides, rather than the homepage, is the tell.
- Volatility: an assistant's answer to the same question changes week to week, and so does the citation. Expect the segment to be lumpy.
What gets cited
- A specific, attributable claim. "The X is the best bike light for commuting because its cut-off beam doesn't blind oncoming traffic and it runs 4 hours at mid mode." The assistant can quote that and name you.
- A stated method. "Compared on specifications, owner review patterns and current price" tells the engine what kind of source the page is, and the engines weight sources that say.
- A named author and a site that says what it is. The E-E-A-T post covers this for Google; the assistants read the same signals, often more literally.
- Current data. Live retrieval means an assistant can see that your price or pick is stale. Pages refreshed from the API get cited over pages with 2024 prices.
- Comparative structure. A table and a "pick this if" per product give the engine a sentence per reader to quote.
- Crawlability. The engines use their own bots (OAI-SearchBot, PerplexityBot and others). Blocking them in robots.txt removes the site from the answers; check what the site's robots.txt and any bot-blocking plugin do.
What doesn't
- Hedged verdicts. "It depends on your needs" gives the engine nothing to attribute, and it quotes the competitor who said which.
- Thin lists. Ten products with a sentence each and no claims; nothing in it is a source.
- Stale data. A guide whose picks are discontinued is visibly wrong to an engine retrieving live.
- Q&A dumps and "AI summaries" bolted onto pages, which the engines don't need and readers skip.
- Special markup. There's no schema that earns a citation; the engines read the text.
Seeing it in analytics
- In GA4, Reports → Acquisition → Traffic acquisition, filter session source for chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com, claude.ai and bing.com/chat. Save it as a segment or an exploration.
- Add a comparison of direct traffic landing on guide URLs (not the homepage) over time; a rise there alongside the segment is the hidden share.
- Read it monthly. The numbers are small enough that a daily view is noise, and the rank tracking post's argument about 28-day windows applies here too.
- Look at which pages the segment lands on; they're the ones whose verdicts are quotable, and a model for the rest.
Whether to change anything
The features that get cited are a clear verdict with a reason, a stated method, a named author, current data, comparative structure and crawlability. That's the list for ranking on Google, for converting the reader who lands, and for surviving a quality review, and nothing on it is specific to assistants.
The only two actions that are specific: check that the engines' bots aren't blocked, and watch the referrer segment. Everything else is "write the honest guide and keep it current", which was already the job. Sites that build a separate AI-optimisation layer are spending on a channel that rewards the same page as every other channel.
Doing it with AffSERP
Guides written by AffSERP's X Best Review tool open with a verdict that names the pick, the reader and the reason in one attributable sentence, state the method as specifications, review patterns and price, and carry a "pick this if" per product and a comparison table, which is the quotable structure the assistants pull from.
The Refresh option keeps picks and prices current from Amazon's API, which matters more for engines retrieving live than for an index. The author is whoever you set in WordPress and the robots.txt is yours to check; the content side of the citation list is what the tool writes by default.
FAQ
Do ChatGPT and Perplexity send traffic to affiliate sites?
Some, and more each quarter, though still a small share of search referrals for most product sites: typically low single digits of what Google sends, with Perplexity the most consistent citer, ChatGPT's search mode growing fastest, and Copilot and Gemini behind. The visitors who do arrive are far down the funnel; they asked a specific product question, got an answer that cited your page, and clicked to verify or to buy. Conversion on those visits tends to be high, which is why the share matters more than its size suggests.
How do I get my site cited by AI search engines?
By being the page an assistant can quote for a specific claim. The engines pull from pages they can crawl, that answer the question in a clearly attributable sentence, that state a method and an author, and that carry current data. A guide whose verdict says 'the X is the best commuter light because of its cut-off beam and 4-hour runtime at mid mode' is quotable; one that says 'there are many great options' isn't. There's no submission process and no separate markup that earns a citation; the features that get cited are the ones an honest guide has anyway.
How can I see AI search traffic in Google Analytics?
Filter the traffic acquisition report by session source and look for chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com and claude.ai; most pass a referrer. Some assistant traffic arrives with no referrer and shows as direct, so a rise in direct visits landing on deep guide pages (rather than the homepage) is a second signal. Build a saved segment for the known referrers and watch it monthly; it's small enough that daily numbers are noise.
Should I optimise my content for AI search?
Not as a separate activity. The pages that get cited are the ones with a clear verdict, a stated method, specific comparative claims, a named author and current data, which is the same list that ranks on Google and converts readers. Chasing 'AI optimisation' tactics (keyword-stuffed summaries, Q&A dumps, special markup) adds nothing an assistant uses and degrades the page for people. Make the honest guide, make it crawlable, keep it current, and check the referrer segment once a month.
AffSERP guides open with one attributable sentence naming the pick, the reader and the reason, state the method plainly, and stay current through Refresh, which is what live-retrieving engines cite.
Start with free words