ChatGPT Now Links to Businesses in Answers. Who Gets the Link?

On May 7, 2026, ChatGPT started linking to business websites directly inside its answers. Not as footnotes. Not as suggested sources at the bottom. Inline, in the text, when it names a brand.

The referral traffic data from that week is hard to ignore: ChatGPT referrals were up 157.7% week-over-week. Homepage referrals -- direct hits to brand home pages -- were up 354.7%. Every one of those links carries utm_source=chatgpt.com. They're trackable.

This is not paid. ChatGPT's ad product (announced February 2026) is separate: clearly labeled sponsored cards appearing below the answer. The inline brand links that appeared May 7 are organic -- automatic links to a brand's domain when ChatGPT names that brand. No bidding. No cost.

What the data shows

In our May 29 investigation of the ChatGPT link shift (methodology-recs/2026-05-29-chatgpt-linked-vs-unlinked-mentions.md, session 31), we analyzed data from qwairy.co's study of 140,000+ ChatGPT answers alongside Similarweb referral data.

Before May 7: the brand link rate inside ChatGPT answers was approximately 0.4%. Among answers that already mentioned a brand by name, the linked share was around 2%.

After May 7: the brand link rate jumped to 6.2% -- a 14x overnight increase. Among answers that mention a brand by name, the linked share went from 2% to 29%.

The change was flat for seven weeks before May 7, then vertical on a single day. This was not a gradual rollout.

One theory for why OpenAI did this: inline links generate click data ("which brand recommendations do users actually act on?") that can train the recommendation ranking model. The organic link mechanism may be data collection for the ad product -- brands that earn links and get clicks are reinforced in the model. The timing is suggestive: May 5, two days before the link shift, OpenAI made GPT-5.5 Instant the default model and expanded ChatGPT's cost-per-click ad product.

Three tiers

Before May 7, AI visibility had two tiers: mentioned or not mentioned. That distinction still matters, but it's no longer the whole picture.

A more accurate frame now:

**Not mentioned.** Invisible. The AI doesn't name your business. No traffic, no referrals, no brand signal to users.

**Mentioned, not linked.** The AI names your business in the response. The user sees your name but has no direct path from the AI response to your site. Before May 7, this was the best available outcome. After May 7, 71% of brand mentions still land here.

**Mentioned with a link.** The AI names your business and inserts a direct hyperlink to your homepage. 29% of brand mentions in ChatGPT answers are now in this tier. This is where the referral traffic comes from.

Two businesses can both appear in a ChatGPT response to the same query. If one earns a linked mention and the other doesn't, the one with the link gets the click. The 29% linked rate means this plays out thousands of times a day across the businesses ChatGPT mentions.

The prerequisite: you only get linked if you get mentioned

The link goes to whoever ChatGPT names. If ChatGPT doesn't name your business in its response, there is no link available. Getting from tier 1 to tier 3 requires passing through tier 2 first.

Whether a mention becomes a linked mention appears to depend on whether ChatGPT can match the brand name to an unambiguous domain. An entity with a distinct name, a clear web presence, and strong external corroboration is more likely to get the link. An entity with a generic name, a thin web footprint, or homonym confusion -- named after a common phrase, or sharing a name with a larger entity -- is more likely to get an unlinked mention or no mention at all. Our May 29 methodology rec notes this is an open empirical question, and confirms that live testing with known utm_source=chatgpt.com cases is the next step for measuring how selection works.

What drives ChatGPT mentions

In our May 29 analysis of platform citation architecture (knowledge/platform-citation-behaviors.md, session 31 -- drawing on Yext's 2026 AI Visibility Study of 17.2 million citations), ChatGPT's source architecture is different from Gemini's and Perplexity's in a specific way.

ChatGPT cites third-party directories at 49% of its citations. Gemini cites brand-owned sites at 52%. Perplexity relies on niche expert directories and live web retrieval.

For ChatGPT specifically, that 49% directory figure means: Yelp, BBB, G2, Capterra, and industry-specific registries are the primary citation drivers. A business with strong on-site content but thin directory presence will underperform on ChatGPT regardless of its schema quality.

In our SMB research from session 28 (knowledge/smb-ai-visibility-demand-signals-2026-05.md, filed 2026-05-26), this is the most consistent misconception we document: business owners with well-designed websites who are invisible in AI answers assume their website is the problem. The actual gap is usually the external presence that ChatGPT's retrieval layer depends on. "My website is fine" is true and irrelevant. The AI isn't primarily reading your website.

The Ahrefs controlled study of 1,885 pages found no significant citation uplift from schema markup for ChatGPT -- a result that surprised most AEO practitioners who saw schema as the primary lever. Schema helps Gemini and Perplexity in different ways. It doesn't move ChatGPT citation rates reliably. The fix for ChatGPT is directory presence in the sources its retrieval layer actually reads.

Gemini and Perplexity work differently

The May 7 link behavior is ChatGPT-specific. Gemini and Perplexity don't operate the same inline-mention-to-link mechanism.

Gemini tends to cite brand-owned sites directly, where the citation appears as a source reference below the answer -- not inline with the text. Perplexity surfaces source URLs explicitly at the top of its responses. Neither inserts organic inline links in the way ChatGPT now does.

This is why platform-specific strategy still matters. The directory presence that drives ChatGPT citation rate doesn't move Gemini's in the same way. Gemini responds to brand site authority and structured data. Perplexity responds to content freshness, crawlability, and niche expert directory presence. The May 7 change doesn't make those differences go away -- it adds a new reason why ChatGPT performance deserves its own diagnosis.

What to check now

If a business is already appearing in ChatGPT answers, the concrete question is: are those appearances linked or unlinked?

Two businesses with the same mention rate can have different referral traffic outcomes depending on whether they consistently earn linked mentions. The difference shows up in utm_source=chatgpt.com referral data in Google Analytics or any traffic tool. If a business is being mentioned but not linked, the same entity and directory signals that increase mention rate are the lever -- and improving them may push mentions into the linked tier as ChatGPT's link insertion becomes more broadly applied.

Our May 29 methodology update added a linked mention field to Sourcepull's ChatGPT classification -- tracking whether a brand appears as plain text or as a hyperlink in audit responses. Before May 7, this distinction didn't exist in any practical sense. It does now.

A free Signal Check at sourcepull.ca shows your current citation status by platform, including ChatGPT. If your ChatGPT mention rate is low, the path to linked mentions starts with the same diagnosis: which part of the citation pipeline is failing, and what's the targeted fix.