ChatGPT now drives 87% of all AI referral traffic. The brands showing up in the answers are not necessarily the ones with the biggest SEO budgets. Here is what is actually working.
AI search referral traffic is still under 1 percent of total pageviews for most publishers, but it is growing fast and ChatGPT alone now accounts for 87.4 percent of it. The brands getting cited are using a specific recipe: technical access via robots.txt and llms.txt, question-based H2s with 1-2 sentence answers, third-party authority on Reddit and G2, and patient consistency. None of it is magic. It is a pragmatic content discipline most marketing teams can execute on this quarter.
Let me get the honest part out of the way first. AI search referral traffic is still tiny in absolute terms. Chartbeat’s March 2026 analysis put it at less than 1 percent of total pageviews across the publishers they track.[1] If you are running a marketing function and someone in your leadership meeting asks why this is on the agenda, that number gets thrown back at you fast.
The reason it is on the agenda anyway: the trajectory. ChatGPT alone now drives 87.4 percent of all AI referral traffic, and overall pageviews from AI sources have grown more than 200 percent year over year.[1][2] Google traffic to publishers has been declining at the same time, with smaller publishers reportedly losing as much as 60 percent of their referral traffic in the AI era.[3] The shift is not happening in a year. It is happening over three to five. The teams optimising now are getting their content into ChatGPT’s answer set while the field is still relatively uncrowded.
Honestly, most marketing teams I talk to are still treating AEO (answer engine optimisation) as something to “look into next quarter.” If you have any kind of consideration-funnel content (comparison guides, “best of” lists, how-tos, definitions), getting that content into AI answers is going to be the difference between being on the shortlist and never appearing at all. People are not going to click through 10 search results and then come to you. They are going to ask ChatGPT, get three names, and that is the shortlist.
You do not need to abandon SEO. Most of what gets you cited by AI search engines also helps you in regular search. The work compounds. The framing just shifts from “ranking” to “being the answer.”
The single most common reason a brand gets zero AI citations is that AI crawlers cannot access the site at all.[4] This is the lowest-hanging fruit and the one I see senior marketing directors miss most often, because it is “an SEO thing” or “an IT thing.” It is your thing. Spend 20 minutes on it this week.
Three checks, in order:
1. Open your robots.txt file. The URL is yourbrand.com/robots.txt. Check whether it is blocking any of these user agents: GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Bytespider. If any of those are in a Disallow line, the relevant AI is currently locked out of your site. Talk to whoever owns your technical SEO and unblock them deliberately. You can choose to keep some out (some publishers are doing this for licensing reasons) but make it a decision, not an accident.
2. Add an llms.txt file. This is an emerging standard, similar to robots.txt, that tells AI models what your site is about, what your products do, and how to reference you.[4] It lives at yourbrand.com/llms.txt. It is a plain markdown file describing your site, your key offerings, your locations, your standout content. Brands using it are seeing better citation accuracy because the AI does not have to guess what you do. Treat it as a 1-page brand brief written for machines.
3. Add structured data (schema) to your most important pages. Article schema for blog posts. FAQ schema for FAQ pages. Product schema for product pages. Organisation schema for your homepage. AI engines extract from structured data more reliably than from unstructured HTML.
None of this is glamorous. All of it is necessary. If your team already has technical SEO in motion, this is mostly a “make sure it includes AI crawlers” conversation. If you do not, it is a single half-day project for a developer.
Once the AI can crawl you, the next problem is whether your content is actually extractable as an answer. AI engines do not rank pages the way Google does. They pull specific passages that answer specific questions and cite those.[5] Your job is to make those passages obvious.
The pattern that works: question-based H2 headings, followed by a 1 to 2 sentence direct answer, then the supporting context.[6]
Concrete example. A typical “good” SEO heading:
“The Benefits of Email Marketing Automation” followed by a 6-paragraph essay.
The AEO-friendly version:
“What is the biggest benefit of email marketing automation?” followed immediately by: “The biggest benefit is segmentation at scale: automated campaigns can deliver personalised content to thousands of subscribers based on their behaviour, which drives an average 41 percent revenue lift compared to one-size-fits-all sends.” Then your 6 paragraphs.
The first 2 sentences after a question heading are doing the heavy lifting. They are what gets pulled into the answer. Write them as if they are the only thing the reader (or the model) will see.
The “perplexity” trick. Surfer SEO’s analysis of LLM citations found that subjective phrases like “I think,” “we believe,” and “in our opinion” actually reduce the likelihood of your content being cited.[7] Models prefer declarative, objective sentences because they have lower uncertainty (in technical terms, lower perplexity). For your AEO-targeted content, drop the hedge words. Lead with the answer, support with the evidence.
What this looks like in practice for the average B2B blog post: rewrite your top 5 traffic-driving articles to include a question-format H2 in the first scroll, with a 2-sentence answer immediately below. That is the highest-leverage content rewrite you can do this quarter.
Here is the part that throws most marketing teams. Your own content is necessary but not sufficient. Different AI engines weight different sources, and the weighting often favours places you do not control.
ChatGPT leans heavily on traditional editorial authority: major publications, Wikipedia, and Reddit.[5] Perplexity is even more skewed: roughly 47 percent of its top citations come from Reddit, with meaningful shares from YouTube, G2, Yelp, and TripAdvisor.[5] If your brand is not being talked about on those platforms, you are missing the second-biggest source of AI citations.
What this means tactically:
Reddit is now a marketing channel. Not for spam. For genuine presence. Find the 3 to 5 subreddits where your customers hang out. Have your subject-matter experts (not your social media manager) participate authentically. Answer questions. Share useful threads. The goal is not engagement metrics. It is being mentioned in a context the AI can find.
Get real reviews on G2, Capterra, Gartner Peer Insights. If you are B2B, this matters more than almost any other AEO move. A real review platform with named reviewers and recent dates is gold for citation engines because they are explicit “consensus” data.
Get into earned media that AI models trust. A guest post on a niche blog with low domain authority does almost nothing. A quote in a TechCrunch, a Forbes article, an HBR piece, or a major industry trade publication does a lot. The data is consistent: high-traffic, high-authority sites earn 3 times more AI citations than low-traffic ones.[8]
YouTube matters more than most marketers expect. Especially for Perplexity citations. If you have an executive who can do a 7-minute talk-to-camera video on your category, get it on YouTube with a transcript. The transcript is what the AI actually uses.
If you do not measure this, you will not improve at it. The good news: the audit is fast. Once you have it set up, it takes about 30 minutes a week.
The routine I run with our clients:
Monday morning, run 5 category prompts across 4 engines. The four are ChatGPT, Perplexity, Claude, and Gemini. The five prompts are the questions your customers actually ask before they buy. For example, if you are a B2B email marketing platform, your prompts might be:
For each, log: Did your brand appear? In what position (first, middle, last)? What competitors were mentioned? What sources did the AI cite?
Track this in a simple Google Sheet. 5 prompts x 4 engines x weekly = 20 data points a week. After 6 weeks, you have a clear picture of where you appear, where you do not, and which engines are easier wins.
For brands that want to skip the manual work, tools like Profound, Athena, and Otterly track AI citations automatically. They are useful, but I genuinely recommend doing the manual version for the first month, because you start to notice patterns the dashboards do not flag (which competitors are everywhere, what kind of language the AI uses to describe your category, what’s missing from your own content).
The biggest mistake I see is treating “AI search” as one channel. ChatGPT, Perplexity, and Gemini behave differently and reward different things.
ChatGPT rewards traditional authority. To increase your citation rate here, focus on getting featured in major publications, having a robust Wikipedia presence (if you are big enough to qualify), and being mentioned regularly on Reddit. ChatGPT pulls heavily from training data updates, which happen on a slower schedule, so consistency over months matters more than recency.[5]
Perplexity does live web searches, so recency matters. Recent blog posts, recent reviews, recent Reddit threads all contribute. If your content strategy is “evergreen, update once a year,” you are leaving Perplexity citations on the table. Add a quarterly refresh cycle to your top 10 most important pages.
Gemini is the fastest-growing AI search source (388 percent growth in late 2025).[1] It pulls heavily from Google’s index, which means traditional SEO wins still apply, plus the AEO-specific moves above. If you are doing SEO well, Gemini is the easiest win.
The implication: split your reporting by engine. “We appear in ChatGPT 4 out of 5 prompts but only Perplexity 1 out of 5” tells you exactly what to work on. “We appear in AI search 25 percent of the time” tells you nothing actionable.
A few tactics getting heavy LinkedIn promotion right now that are not actually moving the needle:
“Stuffing your content with FAQ schema” without substance. Schema helps if the content is genuinely answering questions. Bolting it onto a page that does not actually have clear Q&A in the body does nothing. Worse, some teams are using schema to claim FAQ content that is not on the page, which can earn you a manual penalty in Google.
Paying for AI citation services that “guarantee” placements. Most of these are doing two things: getting you mentioned on low-quality SEO networks, and gaming public Reddit subs in ways that get caught. Pay attention to the long-term reputation cost.
Optimising for AI search by abandoning humans. Some content I have read recently has been so structured for extractability that no human would actually read it. That defeats the purpose. The same content needs to drive trust when a real prospect lands on it. Write for both.
For more on the broader AEO and search shift, our Google AI Overviews paid search analysis covers the paid side, and our LinkedIn AI search guide covers the LinkedIn-specific play.
Three actions, in order:
Monday: Pull up your robots.txt. Confirm GPTBot, ClaudeBot, PerplexityBot, and Google-Extended are not blocked. If they are, talk to your dev team and unblock the ones you want to be cited by.
Tuesday or Wednesday: Pick your top 3 pages by traffic. Rewrite the first H2 on each as a question, with a 2-sentence direct answer immediately below. Time it. It will take less than an hour total.
Friday: Run the 5-prompt audit across ChatGPT, Perplexity, Claude, and Gemini. Log the results. Set a calendar reminder to repeat next week.
Six weeks of this and you will have a real picture of where you sit, what is shifting, and where the gaps are. That puts you ahead of 90 percent of marketing teams in your category.
It means ChatGPT references your website or brand by name in its answer to a user’s question, often with a clickable link. Citations are how AI search engines surface specific sources rather than just generating an answer in their own words.
As of early 2026, AI search referrals are less than 1 percent of total publisher traffic, but ChatGPT alone drives 87.4 percent of that AI referral traffic, and overall pageviews from AI sources have grown more than 200 percent year over year. The volume is small now but rising fast.
Most brands should not block them. If you block them in your robots.txt file, you are taking yourself out of the citation pool. Some publishers block AI crawlers for licensing reasons, but for marketing teams trying to build awareness through AI search, blocking is the wrong default.
SEO optimises a page to rank in a list of search results that a user clicks through. AEO (answer engine optimisation) optimises a passage of content to be extracted as part of an AI-generated answer. AEO favours question-based headings, direct 2-sentence answers, structured data, and authority signals like third-party reviews.
Both, in the same content. The good news is that the patterns that work for AI extraction (clear questions, direct answers, structured layout, real expertise) also tend to make content easier for humans to scan. If you have to choose, choose humans, because the AI is still a referral source. The conversion happens with the human reader.
This guide is part of Future Factors AI’s ongoing effort to make AI useful for non-technical professionals. Written by Hina Mian, Co-Founder of Future Factors AI, an AI training company that has helped 2,000+ learners build practical AI skills through bootcamps, corporate workshops, and keynote sessions. Visit our AI Courses page to learn more.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.
