How AI search shifts the branded/unbranded query split

The branded versus unbranded split has been a stable lens on B2B marketing performance for years. Branded searches tell you whether your demand generation is working. Unbranded searches tell you whether your category content is. The split rarely moved more than a few points quarter on quarter, and most teams used it as a stable backdrop for measuring everything else.

That stability is gone. AI search is reshaping how each query type behaves and the split itself is becoming a less reliable proxy for the underlying picture. This post is what we have actually seen across our client data and what it means for how you measure performance.

The shift in plain terms

Three patterns we are seeing consistently:

Branded search volume is going up in absolute terms. Prospects who have heard about you somewhere, often in an LLM answer, now go straight to a branded query to verify and explore. That used to happen via direct navigation. It is now showing up as branded search.

Unbranded search volume is going down. Some of what used to be unbranded discovery is now happening inside ChatGPT, Claude, Gemini and Copilot, and never reaches Google or Bing as a query. The model answers, the user clicks one or two sources, and the rest of the unbranded discovery journey disappears from your analytics.

The branded/unbranded ratio is shifting upward without the underlying business shifting. Teams looking at the ratio in isolation are concluding that brand is doing better and category is doing worse. In our experience, that is often not what is happening. The ratio is being distorted by where the queries now occur.

If you want background on tracking the underlying behaviour, our piece on tracking AI search traffic covers the analytics setup.

Why unbranded queries are migrating into LLMs

Unbranded queries are mostly research queries. “Best ”, ” for ”, “How does work”. These are exactly the questions LLMs answer well. The user gets a synthesised answer with a few sources, decides which to look at and goes from there.

Branded queries are mostly verification or transactional queries. ” reviews”, ” pricing”, ” login”. These do not benefit as much from synthesis because the user already knows what they want, so they skip the LLM and go straight to search.

The asymmetry is the point. The query types that move into LLMs are unbranded. The query types that stay on Google are branded. The natural consequence is the ratio shifts even if your underlying business is stable.

For the wider strategic context, our piece on branded versus non-branded SEO covers the SEO-side implications.

What this means for measurement

Three measurement adjustments we have made with clients:

Stop reading the ratio as a brand health proxy on its own. Look at branded volume in absolute terms instead. If branded queries are growing, your demand generation is probably working, regardless of what the unbranded line is doing.

Add LLM citation share as a parallel metric to unbranded SEO performance. If your unbranded SEO traffic is flat but your citation share is growing, you are probably winning the category surface in the new place it lives. That has to count.

Track the source mix on branded queries. Are the new branded queries from people who saw you in an LLM answer, in a podcast or in a peer recommendation? Sales discovery notes are the cleanest place to capture this, and the data is messier than you would like, but it is better than nothing.

We are also recommending teams report “discoverability” as a combined metric. Unbranded search traffic plus LLM citation share plus relevant referral traffic from community sources. No single number does the job any more, and pretending one does makes you make worse decisions.

How AI search affects the funnel shape

The classic funnel had unbranded discovery at the top, branded research in the middle and direct conversion at the bottom. The new shape is muddier.

A buyer might:

1. Ask Claude “Best tools for ” and read three vendor mentions
2. Ask Claude “How does compare to ” and form a preference
3. Type "" into Google to find pricing and download a case study
4. Ask their network on Slack what they think of
5. Visit ‘s site directly and convert

In that journey, none of the unbranded discovery touches your analytics. Two branded interactions touch it. The conversion looks like it came from branded direct, when in fact most of the work was unbranded discovery in an LLM you cannot see.

The implication is that branded performance is increasingly the visible proxy for invisible category performance. If your branded conversions are growing but your unbranded SEO is flat, that is not a contradiction. It is the new shape.

What to optimise differently

Five practical adjustments we are making in client programmes.

Treat your name as a category synonym. When someone asks an LLM “Best alternatives to ”, you want the model to know your brand is in that set. That requires named-entity presence in third-party content and clear positioning on your own site. Our piece on ChatGPT keeps recommending your competitor covers the displacement playbook.

Make branded landing experiences AI-aware. A prospect who arrives via branded search after seeing you in an LLM answer has different expectations from one who arrived from organic. They have heard a summary already, possibly an inaccurate one. The page needs to reinforce the right framing fast.

Invest in comparison and category content even if the unbranded SEO traffic is declining. That content is still being read, just inside the LLM. It still drives consideration. The fact that you cannot see the visit does not mean the visit is not happening. Our piece on comparison content that ranks is still relevant, and our forthcoming piece on optimising for compare X to Y prompts covers the AI-specific structure.

Track citation share at category level. For each unbranded prompt cluster, how often are you cited across at least four LLM surfaces. This is the closest proxy to “are we visible in the new category research layer”. Our walkthrough on tracking AI citations through Profound versus manual prompt audits covers the tooling.

Update your reporting to senior stakeholders. If you are still presenting an unbranded-traffic-as-category-health story, you are inviting the wrong conversation. Re-shape the narrative around discoverability, and explain why.

The bits we cannot prove cleanly yet

Several uncertainties we are honest about with clients.

We do not have a clean view of how much of unbranded research has migrated into LLMs versus simply declined as buyers behave differently. Both effects are real. The split between them is hard.

We cannot reliably attribute a branded query to a specific LLM citation, except via sales discovery notes. The data is partial.

We do not have a stable longitudinal picture, because the LLM surfaces themselves change frequently and the citation behaviour changes with them. What we measure this quarter may not match next quarter.

The honest position is that the branded/unbranded split is becoming less informative as a single metric, and the teams who notice fastest will adjust their reporting and their investment ahead of the teams who do not.

A note on senior buy-in

This is not always an easy conversation with a CFO or CEO who is used to reading the unbranded number as a category indicator. We find it helps to show, side by side, the unbranded SEO trend, the LLM citation share trend and the branded conversion trend over the same period. If two of three are flat or up, the discoverability story is intact even if the unbranded traffic line is sliding. The goal is not to defend the SEO budget. It is to redirect it where the buyers now are.

For wider context, our content KPIs for the AI search era piece covers the broader measurement picture.

If you’d like a second opinion on how to measure your AI search performance honestly, drop us a line. You can also see how we approach this work on our AI SEO services page or our SEO services for the wider foundation.