GEO for Affiliate Sites: How to Get Cited by ChatGPT, Google AI Overviews and AI Mode, Which AI Crawlers to Allow and How to Measure AI Traffic

GEO, or generative engine optimization, is the art of getting your pages quoted and linked inside AI answers: ChatGPT search, Google AI Overviews, AI Mode, Perplexity, Copilot. Half the internet is selling it as a brand-new discipline with a brand-new price tag. My honest take after years of SEO on affiliate sites: it's about 80% plain good SEO, and the other 20% is a writing discipline most affiliate sites never bothered with. That 20% is where the edge is.

This guide is written for people who run affiliate and content sites, not for agencies with a client deck to fill. We'll look at how AI answers actually choose their sources, what Google officially says (it's shorter than the gurus would like), which crawlers to let in, how to write pages a machine wants to quote, what llms.txt really does, how to see AI traffic in your analytics, and why the August 2026 spam update is a big red sign over "GEO at scale".

What GEO actually is (and why it's mostly SEO with better manners)

GEO means shaping your content so that AI search systems find it, trust it and cite it with a link when they write an answer. The term comes from a research paper, "GEO: Generative Engine Optimization" by researchers from Princeton and IIT Delhi, presented at the KDD conference in 2024. You'll also see AEO (answer engine optimization) and LLMO. Same idea, different badge.

That paper is worth knowing because it tested what actually moves the needle. Adding quotations, statistics and cited sources made a page noticeably more visible in generated answers, with gains of up to about 40% in their setup. Keyword stuffing, the old SEO reflex, did not help at all and scored below doing nothing. Read that again if you still write "best cheap VPN cheap" three times per paragraph.

So the shift is not "forget SEO": AI systems still retrieve your page through search indexes first. The shift is in how you write once you're found: liftable facts, numbers next to names, answers before stories. Ironically, the robots are pushing affiliate sites toward what readers wanted all along.

How AI answers pick their sources: retrieval and query fan-out

AI answers don't "know" your site. They search, read a handful of retrieved pages, and write an answer from them. The key mechanism is query fan-out: the system splits one question into several narrower searches, runs them, and collects supporting pages for each piece.

Both big players describe it openly. Google's documentation for site owners says AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources, and that this lets them show a wider and more diverse set of links than classic results. OpenAI's help center says ChatGPT search typically rewrites your question into one or more targeted queries, sends them to its search providers, and may follow up with more specific queries after reviewing the first results.

What this means for an affiliate site in practice:

• You don't need to rank for the big head term. If someone asks "which digital product platform takes the smallest cut for a $20 ebook", the fan-out may search for each platform's fees separately. A page that states one platform's fee clearly can get pulled in, even if it never ranks for "best platform to sell ebooks".
• Specific beats general. Sub-queries are narrow and often contain exact numbers or names. Pages with exact, current facts match them; pages with "affordable pricing" don't.
• Each section competes on its own. The machine may lift one paragraph or one table row, so every block should make sense without the rest of the page.

What I spotted in onlinemm.org's own Search Console

This isn't theory for me. Going through onlinemm.org's query report in Search Console, I noticed strange strings landing on our pages about selling digital products, things like "+site.gumroad.com pricing 10% 0.50" and a dozen variants with "monthly fee", "direct sales" or "digital products" tacked on. No human types that. It looks exactly like a machine double-checking a fact: "is it true that Gumroad charges 10% plus $0.50?"

Our pages showed up for those strings because they state that fee plainly, in one sentence, next to the platform's name. Long question-style queries like "where can affiliates access high-quality traffic without broker fees?" land on our traffic articles the same way. The numbers are small, so I won't wave them around, but it's a real window: you can see fan-out sub-queries in your own data.

What Google officially says about AI features and your website

Google's position fits in one line: there are no additional requirements and no special optimizations needed to appear in AI Overviews or AI Mode. A page has to be indexed and eligible to show in Search with a snippet, and that's the technical bar. Here's how the popular claims stack up against Google's own guide for site owners, "AI features and your website":

One detail from that guide matters a lot for affiliates: nosnippet, data-nosnippet, max-snippet and noindex limit what Google can show from your pages, AI features included. Some plugins and "content protection" tweaks set these quietly, and the owner never finds out why the site isn't cited. Beyond that, Google's best practices are the boring basics: let crawlers in through robots.txt and your CDN, link internally, keep important content as text.

Is AI search killing affiliate clicks?

It's cutting them on some queries, and the best independent data shows by how much. In a Pew Research Center study of real browsing by about 900 US adults, people clicked a traditional search result on roughly 8% of visits when an AI summary appeared, versus about 15% when it didn't. Links inside the summary itself got clicked on only about 1% of visits.

The same study has two more numbers I find even more useful. An AI summary showed up on nearly one in five Google searches overall, but on more than half of searches with ten or more words, and on about 60% of searches phrased as questions starting with who, what, when or why. And users ended their browsing session more often after a page with a summary (about a quarter of visits) than after a page without one.

Translation for affiliates: long, question-style informational queries are where the summaries live. Short commercial queries ("[product] review", "[tool] pricing", "[brand] coupon") are hit less. Here's a rough example of what that means for one page, just to make the scale tangible.

Example, not a real result: say a review page gets 2,000 visits a month from Google, and half of them come from queries that start showing an AI summary. If clicks on those queries drop in the same proportion as in the Pew data (8 out of 15, about 53% of the old level), those 1,000 visits become roughly 533. The page now gets about 1,533 visits, a drop of around 23%. At an earnings per click of $0.40, monthly revenue goes from $800 to about $613.

That's painful, but it's not "the end of affiliate sites". And it tells you where to work: protect the commercial pages, and make the informational pages earn citations instead of fighting a battle for clicks they'll partly lose anyway. If you want to run this math on your own numbers, the earnings calculator does it in a minute.

Robots.txt: which AI crawlers to allow and which to block

Short answer: allow the search and user-fetch bots, decide separately about the training bots. Most AI companies now split their crawlers by purpose, so you can stay visible in AI search while refusing to feed model training.

OpenAI states that each of its robots.txt settings is independent: you can allow OAI-SearchBot and disallow GPTBot, and changes for search take roughly a day to be picked up. A typical setup for an affiliate site that wants AI citations but not training looks like this, as lines in robots.txt:

• User-agent: GPTBot / Disallow: /
• User-agent: ClaudeBot / Disallow: /
• User-agent: OAI-SearchBot / Allow: /
• User-agent: Claude-SearchBot / Allow: /
• Googlebot, Bingbot and PerplexityBot: leave them alone (no Disallow)

Blocking training bots is a matter of principle more than money: training sends no visitors either way. What you must not do is block the search bots by accident.

And "by accident" happens more than you'd think. Cloudflare switched to blocking AI crawlers by default for new domains on July 1, 2025, asking at sign-up whether to allow them, and some hosting firewalls and security plugins have similar toggles. It's a classic mistake: weeks spent "optimizing for ChatGPT" on a site the CDN is politely turning bots away from. Check your CDN's bot settings once, and if you run your own server (a few of my projects live on an InterServer VPS), grep the access logs for OAI-SearchBot and Claude-SearchBot to confirm they actually get 200 responses.

If you're curious how the same robots.txt logic works for SEO tools rather than AI, I covered blocking Ahrefs, Semrush and Majestic in how to hide backlinks from competitors. Same file, very different stakes.

How to write pages that AI wants to quote

Write every section so that its first two sentences answer the question in the heading, with the key name and number inside them. That one habit does more for AI citations than any plugin. The machine is looking for a passage it can lift without editing; give it one.

Here's the checklist I use for onlinemm.org articles, and it works the same for any affiliate page:

1. Short answer on top. 40 to 80 words right under the title that actually answer the main query. No "in this article we will explore".
2. Answer first in every section. Details, caveats and stories come after the answer, not before.
3. Self-contained paragraphs. Don't start a paragraph with "It" or "This" when the subject is three paragraphs up. Name the product. A lifted paragraph has no context.
4. Exact facts next to names. "Gumroad charges 10% + $0.50 per direct sale" is quotable. "Fees are reasonable" isn't. This is exactly what the fan-out sub-queries hunt for.
5. Real HTML tables for comparisons. Prices, fees, payout minimums, features. Tables are easy for machines to read row by row, and easy for humans to scan.
6. Headings that match real questions. Roughly half of them as questions people actually ask, the rest as plain topic labels. Don't turn every heading into a question, it reads like a robot wrote it (which is ironic).
7. Cover the neighbouring questions. What it is, how it works, who it suits, what it costs, limits and risks, mistakes, alternatives. That's the fan-out map; if you cover it, more sub-queries land on you.
8. Define terms in one phrase. First time you write EPC or CPA, explain it. AI answers are written for beginners, and they pick sources that explain.

The classic mistake on affiliate sites is the 300-word intro: how the author discovered the product, a paragraph about "today's fast-paced world", and only then the answer. Put the story later; it's still valuable, just not as a gate.

One more thing that separates a citable affiliate page from a thin one: it says what the product is bad at. AI answers love balanced sources, and so do readers. A "pros and cons" table with real cons, a "who should skip this" paragraph, a note about the plan limits. If you need a refresher on building review and comparison pages that convert, my guide to affiliate money pages goes through the structure.

Original data: the part AI can't make up for you

The single most reliable way to get cited is to publish something nobody else has: your own numbers, tests, measurements or collected data. An AI answer that needs a fact only you published has to cite you. Everything else can be summarized from ten other pages.

You don't need a research budget. Affiliates sit on more original data than they realize:

• Your own tests. Signup flows, how long payout approval took, what the dashboard actually shows, which features are behind a paywall. Screenshots and dates of real events help.
• Price and fee tables you maintain. A table of ten platforms' fees, checked against their official pages, is boring to make and gold to cite. Keep it current; stale numbers get you dropped.
• Calculations. "On a $20 product, platform A leaves you $X and platform B leaves you $Y" is original even when the inputs are public. We did exactly that in our comparison of Gumroad, Payhip and Lemon Squeezy fees, and it's no accident that the fee checks in Search Console landed on that topic.
• Small surveys. Ask your email list or Telegram followers three questions. Fifty answers, honestly described as fifty answers, is a real data point.
• Your affiliate manager. An old trick of mine: your affiliate manager is your friend. He sees full stats across offers, traffic sources and GEOs, and knows what converts right now. Ask whether you can publish aggregated, anonymous numbers (typical approval rates, which countries convert best, how a seasonal promo performed). Many programs are happy to give you a quote or a stat for a review, because it's free publicity for them.

Calculators work too, and they have a bonus: even if an AI answer gives the summary, people still click to plug in their own numbers. That's the logic behind the free tools on this site, like the niche keyword profit estimator. A tool is the one thing a text answer can't replace.

Entity and author signals: making sure the machine knows who you are

AI systems prefer sources they can identify: a clear site, a clear author, a consistent track record across the web. For a small affiliate site that means making your identity boringly consistent, not inflating it.

• A real About page and author page. Who writes, what experience they have, how the site makes money. Same name and bio everywhere.
• Author on every article. A byline linked to the author page, with "Updated" and "Fact-checked" in the header. Our pages do this; it costs nothing.
• Schema that matches what's visible. Article with the author as a Person, Organization on the homepage. Google explicitly asks that structured data match the visible text, so don't mark up an "expert reviewer" who isn't on the page.
• Consistent brand mentions elsewhere. Your site's name next to your topic on other sites, forums, podcasts and social profiles. AI answers lean on how often and where a brand is discussed, not only on links.

About mentions on other sites: some affiliates buy guest posts on link marketplaces such as Serpzilla. It can make sense for getting your brand discussed on relevant sites in your niche and for referral visits. Be honest with yourself about the risk, though: Google's spam policies treat paid links meant to pass ranking credit as link spam, and paid placements should carry a sponsored or nofollow attribute. Buy placements for the audience and the mention, not for PageRank, and choose sites real people read.

For promoting the site without paying for placements at all, my list of less obvious blog marketing tactics covers answering questions where people already ask them, guest appearances and audience swaps. Those are also exactly the places AI systems read when deciding who's an authority.

llms.txt: what it is and what it isn't

llms.txt is a proposed file at the root of your site (yoursite.com/llms.txt) that lists your most important pages in Markdown, meant to help language models find your best content. It was proposed in 2024 by Jeremy Howard of Answer.AI. It is a proposal, not a standard, and that's the whole story in one sentence.

What's actually true:

• Google says you don't need AI text files or new machine-readable files to appear in AI Overviews or AI Mode.
• OpenAI's publisher FAQ about appearing in ChatGPT search talks about allowing OAI-SearchBot and using noindex to opt out. It doesn't mention llms.txt.
• Even Rank Math, which ships an llms.txt generator, describes the format as a proposal rather than an official standard, with limited adoption.
• I haven't seen any of the major AI search providers publicly confirm that llms.txt is used to choose which sources to cite.

Should you have one? If you're on WordPress, the Rank Math plugin has an "LLMS Txt" module: switch it on in the dashboard and edit it under General Settings. Five minutes, zero risk; just don't count it as GEO work. If someone sells you an "llms.txt optimization package", that's the 2026 version of submitting your site to 500 search engines.

How to measure AI referrals in Google Analytics and Search Console

You measure AI search in three places: Search Console for Google's AI features (mixed into normal data), your analytics for referrals from chatbots, and Bing Webmaster Tools for Copilot citations. None of them gives the full picture, together they give a decent one.

1. Search Console, Performance report, "Web". AI Overviews and AI Mode data is included here with no separate filter. Watch pages whose impressions grow while CTR falls: that's often a sign an AI summary is answering the query above you.
2. Filter for fan-out queries. In the Queries tab, add a query filter with "Custom (regex)" and paste ^(\S+\s){9,}\S+$ to show queries of ten words or more. Then try site\. or a plus sign to catch the machine-style strings. This is how I found the Gumroad checks.
3. Google Analytics 4 referrals. In Reports, Acquisition, Traffic acquisition, switch the dimension to "Session source" and filter with a regex such as chatgpt\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai. Save it as a comparison or an exploration so you don't rebuild it every month.
4. ChatGPT's UTM tag. OpenAI says ChatGPT automatically adds utm_source=chatgpt.com to referral links, so those visits show up as a clean source in GA4. Some AI apps and in-app browsers strip the referrer, so a slice of AI visits will hide under "Direct". Don't expect perfect numbers.
5. Bing Webmaster Tools, AI Performance. Free, and underrated. It shows how often your site is cited in Microsoft Copilot and Bing's AI summaries, which pages get cited, and sample "grounding queries" the AI used to find you. It doesn't show clicks, but the grounding queries are a ready-made list of questions to answer better.
6. Track conversions, not visits. Mark your affiliate link clicks as key events in GA4 and compare conversion rate by source. A small number of AI referrals that convert well is worth more than a pile of bounces.

Tag links in your own newsletters and social profiles with UTM parameters so they don't blur into organic AI referrals; our UTM generator builds them without typos.

For checking whether you're cited at scale, manual prompts get old fast. Paid SEO suites such as Semrush and Ahrefs now have AI visibility trackers. A cheaper route for tech-savvy folks: Apify has ready-made scrapers in its store for Google AI Overviews, AI Mode and other AI answers, and its free plan includes $5 of platform credit every month (the Starter plan is $19 a month). Run your 20 money keywords weekly and log which domains get cited. One honest warning: scraping search engines can conflict with their terms of service, so keep volumes modest and treat it as research.

Google's August spam update: why "GEO at scale" is a trap

Google released the August 2026 spam update on August 18 and finished it on August 21, according to its Search Status Dashboard. It applied globally and to all languages. Google didn't publish a blog post explaining what it targeted, but case studies from SEO consultants paint a consistent picture.

The best-documented ones, published by Glenn Gabe of GSQI, describe sites that combined programmatic or mass-produced pages with AI-generated text. One was a thin Amazon affiliate site built from product data with AI-written filler at the bottom of pages; it lost rankings for more than 14,000 queries. Another, a programmatic site in a sensitive money niche, lost over 200,000. He notes that an impacted site can drop across Google's surfaces, AI Overviews and AI Mode included, and at least one of his cases was hit across sections, not just the weak ones.

That's the part GEO sellers skip. The fastest way to "cover all the fan-out sub-queries" is to generate thousands of pages, one per question, one per city, one per product variant. That's precisely what Google's spam policies call scaled content abuse: many pages made mainly to manipulate rankings rather than help people, however they're produced. And pages that wrap affiliate links around copied merchant descriptions fall under thin affiliation. Neither policy is new, but the August update showed they're enforced quickly and that the damage reaches AI features too. Before you scale anything, check how your site looks to a pattern detector: Bot or Built, a free tool I built, scans a domain for AI-builder traces, chatbot-style writing and bursts of pages published at once. It's a probability, not a judgment, but if your own site lights up, a spam classifier may see the same thing.

The fix isn't to avoid AI tools (I use them, and listed the ones worth paying for in AI tools for affiliates). It's to keep a human on the parts that make a page worth citing: facts, tests, data, opinion. One good page answering ten related questions beats ten thin pages answering one each. And if you want your content used by AI agents directly, that's a different game with different money, covered in how to make money with MCP servers.

The fear that makes affiliates do exactly the wrong thing

The fear behind most bad GEO decisions is simple: "AI answers will take all my clicks, so my site is already dead." It pushes people into one of two traps, and both are all over affiliate forums right now.

The first trap is giving up: CTR slides, a headline says "zero-click", publishing stops. The second is the panic sprint: clicks per page are falling, so make ten times more pages with AI. It feels like action, which is why it tempts, and it's exactly the pattern the August update punished.

Look at the actual numbers instead of the headlines. In the Pew data, AI summaries appeared on fewer than one in five searches overall, and the click rate dropped by about half on those, not to zero. Your exposure depends on your query mix, and that's something you can check in an afternoon. Here's how I'd turn the fear into a plan:

• Measure your real exposure. List your top 20 pages by revenue and check, by hand, which of their main queries show an AI Overview. Usually it's fewer than you feared, and the money pages are rarely the worst hit.
• Move effort to where the click survives. Reviews with your own tests, comparison tables, calculators, templates, step-by-step setups with screenshots. People click for things a summary can't give them.
• Diversify the traffic. An email list, a Telegram channel, a YouTube presence. Not because SEO is dying, but because one source was always a risk. Our community on Telegram, @affiliatepath, is new and growing from zero, and that's fine; everyone starts there.
• Keep a test budget, not a panic budget. Rework five pages using the writing checklist above, wait six to eight weeks, compare. Then decide. Small experiments beat a big rewrite done in a hurry.

A GEO plan for an existing affiliate site

If I took over an affiliate site tomorrow, this is the order I'd do things in. The first steps take an afternoon, the rest is steady work.

The bottom line: GEO isn't a new religion, it's a sharper version of honest SEO. Let the right bots in, answer first, put real numbers next to real names, publish something nobody else has, and keep away from page factories. The sites that get cited are, surprise, the ones that were worth reading in the first place.