AI Search Stopped Sending Traffic to Your Blog. It's Sending It to Your Homepage.

For eighteen months the advice was consistent, and it was reasonable: write more content, structure it cleanly, get cited by ChatGPT, collect the clicks. Agencies sold it. Tool vendors built dashboards for it. A lot of businesses paid for a lot of articles on the strength of it.

Then on 7 May 2026, OpenAI changed one thing about how ChatGPT renders an answer, and the destination of the traffic moved.

This post is what actually changed, what the numbers say, what they don't say, and what a business with a normal budget should do about it this quarter. Every figure is attributed. Where the data is thin, we say so.

01 What actually changed on 7 May 2026?

Before that date, when ChatGPT mentioned a company, the company's name appeared as plain bold text. If you wanted to visit the site, you had to notice and click a citation chip — a small, easily ignored source marker usually pointing at a deep page, a review, or a listicle that mentioned the brand.

After that date, ChatGPT began embedding the brand's own homepage URL inline, as a clickable link, inside the sentence naming the brand. The share of ChatGPT responses containing a URL at all jumped from roughly 4.5% to somewhere between 20% and 24% — about a fivefold increase. Referrals to monitored brand sites rose 60–65% essentially overnight and stayed there.

Homepages went from a rounding error in OpenAI's referral mix to roughly one in four clicks, in a week.

Source: Profound, "Is zero-click marketing dead? The branded link update", based on live referral monitoring across a basket of brand websites. The change coincided with OpenAI opening self-serve ads to the general public on the same day.

02 How big was the shift, page type by page type?

A separate study by Botpresso measured 52 websites (9 B2B, 43 B2C) across two matched six-week windows — 26 March to 6 May 2026 against 7 May to 17 June 2026. It is the most granular public breakdown of where the traffic went.

Source: Botpresso, "98% growth in ChatGPT referral traffic after the May 2026 links update", n=52 sites, six weeks pre vs six weeks post.

Homepages moved from 4.4% to 31.2% of all ChatGPT-referred traffic. Landing pages still carry the largest absolute share — 57% after the change — because that is where product and service pages sit. Blogs are the only category that went backwards.

The industry split matters too. Profound found B2B software and SaaS gained over 200%, while e-commerce and retail were essentially flat — which suggests ChatGPT routes company recommendations differently from product recommendations. If you sell a service, this update was aimed at you. If you sell physical products, less so.

03 Before you rebuild everything: how big is this channel, really?

This is the part that gets left out of most posts on the subject, and leaving it out is how people end up spending a quarter's budget on a channel that sends them forty visits.

Contentsquare analysed 99 billion sessions across 6,500 sites for its 2026 benchmark, published 3 April 2026. Its finding: AI-referred traffic is 0.2% of total visits. Growing at +632% year on year, converting at 1.3% (up 55% YoY), with bounce rates about 5% better than the average channel — but 0.2%.

Semrush's counter-argument is that the visits are worth more than they look: it puts the average AI search visitor at 4.4x the value of an average organic search visitor. Treat that number carefully — Semrush has not published the sample size or methodology behind it, and it is a vendor selling AI visibility tooling. Adobe's retail data is more specific and points the same direction: AI-referred visits to retail sites bounce 27% less and last 38% longer.

Sources: Contentsquare 2026 Digital Experience Benchmark; MarTech reporting on Semrush; Adobe Analytics (2025), via Semrush.

04 Why "optimise for ChatGPT" is already outdated advice

Even if you accept the homepage shift, targeting ChatGPT alone is a shrinking bet. Similarweb's platform data for June 2025 to May 2026 shows how quickly the ground moved:

Source: Similarweb, AI search stats 2026. Total AI platform visits reached 9.5 billion monthly, up 70% year on year, across 655 million unique visitors.

And the platforms behave differently. Semrush's 2026 AI Visibility Index — 126 million US prompts analysed between January and April 2026, released 26 June — found that ChatGPT cites an average of 15 sources per response while Gemini cites three. Three. Being one of fifteen is a reasonable target. Being one of three is a different sport.

The same index found only 36 brands held top-100 visibility across all four platforms (ChatGPT, Gemini, Google AI Mode, AI Overviews) for the whole study period. And in the Botpresso data, the top five sites absorbed 91% of the net traffic gains — the median site grew 54%, not 98%. This is a winner-take-most channel.

If you are not already the entity an assistant names, being cited in the footnotes buys you very little.

05 So what is the blog actually for now?

Not clicks. Not from AI search, anyway.

The blog's job has changed from traffic asset to evidence asset. It is the material the model reads to decide whether your brand is a credible answer to a question — and then it links your homepage, not the article. That is a demotion in attribution and a promotion in influence, and it should change what you write.

• Write to be quoted, not to be visited. Self-contained claims with a number and a date in the same sentence get lifted cleanly. Paragraphs that require three prior paragraphs of context do not.
• Cover the questions that precede the purchase, not the ones that follow it. "What should a Meta Ads retainer cost in India" earns a mention. "10 tips for better ad copy" earns nothing.
• Stop measuring articles on sessions alone. If AI referral clicks land on the homepage, your article's contribution shows up as an unattributed homepage visit or a direct visit. Judge content on brand mentions and citations, not just GA4 landing-page reports.
• Publish fewer, denser pieces. Fifteen thin posts give a model fifteen weak signals. One page with real numbers and a clear position gives it one strong one.

06 How do you make a homepage that an AI engine can answer with?

If the homepage is now the destination, it has to do the job a landing page used to. Most B2B and service homepages fail this test because they are written to impress rather than to inform. Run yours against these six checks:

1. Say plainly what you do, for whom, and where — in the first 100 words, in prose, not in a slogan. A model cannot extract "We unlock exponential possibility." It can extract "Tomar Marketing is a growth marketing agency working with D2C, education, real estate and B2B businesses in India."
2. Put prices, or price ranges, on the page. "How much does X cost" is one of the highest-volume prompt shapes in existence, and a page with no number cannot answer it. This is also why price-transparent sites get named more often than premium ones.
3. Add Organization schema with sameAs pointing at every profile you control — LinkedIn, Google Business Profile, Instagram. This is entity disambiguation: it is how a model knows which "Tomar" you are.
4. Keep name, address and contact identical everywhere. Inconsistent entity data is the most common and most boring reason a brand fails to get named.
5. Answer objections on the page. Whether ad spend is included, how fast you can start, whether you work with existing teams. If the answer only exists in a sales call, it cannot be cited.
6. Make sure it renders without JavaScript. Content injected client-side is inconsistently read by crawlers. Server-rendered HTML is not optional here.

None of this is exotic. It is the same work that makes a homepage convert a human being — which is the point. The reason we cover it under web and conversion rather than as a separate "AEO package" is that there is no version of this that helps machines and not people.

07 How do you measure any of this?

Badly, at the moment, and you are not alone: Semrush found 45% of marketing leaders cannot accurately measure their brand's visibility in AI answers, and only 9% have tools covering all the relevant metrics.

What you can do this week without buying anything:

• Build a custom channel group in GA4 for AI referrers. Match source against chatgpt.com, openai.com, perplexity.ai, gemini.google.com, claude.ai, copilot.microsoft.com. Without this they scatter across Referral and Direct and you will never see the trend.
• Segment by landing page. Given the May change, expect homepage-first sessions. If your AI referrals still land mostly on blog URLs, that is worth understanding, not celebrating.
• Track mentions, not just clicks. Run your ten most commercially important prompts across ChatGPT, Gemini and AI Mode monthly, by hand, and log whether you are named. It takes twenty minutes and it is more honest than most dashboards.
• Watch server logs for AI crawlers — GPTBot, OAI-SearchBot, PerplexityBot, Google-Extended. If they are not fetching your pages, nothing downstream can happen.

One structural finding from the Semrush index is worth more than any tool: organisations that ran SEO and AI visibility as one function reported increased traffic or leads 81% of the time. Organisations that ran them as separate workstreams: 36%. That gap is the whole argument for treating this as part of an existing growth engine rather than a new department.

08 What not to do right now

Four things are currently being sold that you should refuse:

• Guaranteed placement in AI answers. Nobody controls this. There is no ad product, no submission form, no index to buy into. Anyone promising it is either lying or describing something else. (OpenAI does now run ads — Similarweb found roughly 26% of ChatGPT responses carry them — but that is a separate, labelled surface, not the organic answer.)
• A 40-article "AEO content programme." The data says blog traffic from ChatGPT declined. Buying volume into a declining destination is the single most expensive mistake available in this category right now.
• Chasing unconfirmed algorithm updates. Third-party trackers registered volatility in early August 2026 and a wave of "August core update" content followed. Google's Search Status Dashboard has confirmed no such update. The most recent confirmed events are the May 2026 core update and the June 2026 spam update. Unconfirmed volatility frequently reverses; content written about it does not.
• Ignoring it entirely because it's 0.2%. The compounding rate is +632% YoY, and the fix is a week of work on pages you already own. The cost of being early here is small. The cost of being late is that a competitor becomes the default named answer in your category, which is not a position you win back cheaply.

09 The short version

ChatGPT now links your homepage inside its answers. Homepage share of its referrals went from about 4% to between 24% and 31%; blog sessions fell 28%. The channel is still only 0.2% of total web traffic, but it is growing at 632% a year and converting well. ChatGPT's own share of AI traffic dropped from 76% to 53% as Gemini and Claude took ground, so single-platform optimisation is already obsolete.

The work that follows from all of that is unglamorous: make your homepage state plainly what you do, for whom, where, and at what price; mark up your entity properly; keep your data consistent; write fewer, denser, more quotable pages; and measure AI referrals separately in GA4 before you spend anything on tooling.

If that sounds like the same advice as "build a page that converts a human being" — yes. That is the finding. The machines have simply raised the penalty for a vague homepage.

Related reading: Where growth leaks: the 8 places small business marketing budgets quietly disappear.

Sources

1. Profound — Is zero-click marketing dead? The branded link update. Live referral monitoring; 7 May 2026 change.
2. Botpresso — 98% growth in ChatGPT referral traffic after the May 2026 links update. n=52 sites, six weeks pre vs post.
3. Similarweb — AI search stats 2026. June 2025 – May 2026 platform traffic.
4. Semrush — 2026 AI Visibility Index. 126M US prompts, Jan–Apr 2026, released 26 June 2026.
5. Contentsquare — AI-referred traffic benchmarks. 99B sessions, 6,500 sites, published 3 April 2026.
6. MarTech — Average LLM visitor worth 4.4x organic search visitors.