Is SEO dead? What AI search changed, and what still works.

No. SEO is not dead, but the thing most companies pay an SEO agency for has changed underneath them. Rankings still decide who gets found. What changed is how many clicks a ranking is worth, and where a buyer's research starts.

If you sign the invoice and don't live in the tools, that distinction is the whole question. This article gives you the short answer, the evidence behind it, and a way to decide what to keep paying for, what to cut and what to start measuring.

Is SEO dead, or just different?

SEO is alive and still the base layer of being found, but it is no longer the whole job. Google's own documentation says the best practices for appearing in Google Search also apply to its AI features, and that there are no additional requirements to appear in AI Overviews or AI Mode. A page that Google can't crawl, index or trust has no chance of being summarized.

So "is SEO being phased out?" has a plain answer. The discipline is not going away, because the engines that answer questions still read the same web. What is being phased out is the idea that a page-one ranking guarantees traffic.

What did AI search actually change for B2B?

Two things changed, and they point in the same direction: fewer clicks per search, and a different place where buying research begins.

Fewer clicks per search

Seer Interactive tracked 3,119 informational queries across 42 client organizations. Organic click-through on queries that trigger an AI Overview fell from 1.76% in June 2024 to 0.61% in September 2025 (Seer Interactive, November 2025). Ahrefs looked at 300,000 keywords from a different angle and found position-one click-through of 0.016 with an AI Overview against 0.039 without (Ahrefs, February 2026).

Two firms with different datasets reached the same place: a top ranking on an informational search now earns roughly 60% fewer clicks when an AI summary sits above it. The wider picture of how AI summaries change click behavior is in GEO vs SEO vs AEO.

One limit on that data: it covers informational queries. A search for a vendor's name or "pricing" behaves differently, and most of your pipeline probably comes from those.

A new first stop for research

G2 surveyed more than 1,000 B2B software buyers in August 2025 and found 51% now start their research in an AI chatbot rather than in Google Search (G2, October 2025). ChatGPT was the favorite, at 47%.

Yet Conductor, measuring 13,770 enterprise websites and 3.3 billion sessions, found AI assistants send only 1.08% of visits, with 87.4% of those from ChatGPT (Conductor, 2025). Those two figures don't contradict each other. G2 counts where a buyer says research begins. Conductor counts clicks that arrive at a site. An assistant that answers the question in the chat sends no click, and the buyer may show up days later as a branded search or a direct visit. That reading is our inference, not something either study measured.

The practical effect is that your pipeline can shift before your traffic report notices. If an assistant helps a buyer build a shortlist, being named in its answers is part of the job, and no click-based report will show you that.

What still works, and why?

What still works is the content and trust that every engine reads: pages that answer a real question better than the next page, and a company that other sites and people mention by name. Google's guidance and the independent data agree on this.

• Crawlable, accurate pages. If your pricing page, product pages and case studies are indexed and say what is true today, both Google and the assistants have something to quote.
• Original substance. A page that only restates what ten others say gives an AI summary no reason to cite you. A page with your own data, your own process or a named customer result does.
• Being mentioned elsewhere. Ahrefs studied 75,000 brands and found branded web mentions correlate with AI Overview visibility at 0.664, against 0.218 for backlinks (Ahrefs, May 2025). That is a correlation, not proof of cause, but it points at reputation more than link counts.
• Ranking still helps, but it no longer decides citations. Ahrefs also found that only about 12% of the URLs cited by ChatGPT, Gemini and Copilot rank in Google's top 10 for the same prompt (Ahrefs, August 2025). The citation-versus-ranking picture keeps moving between studies, which is why we kept the detailed comparison in one place.

Seer's data adds a useful wrinkle. Brands cited inside an AI Overview earned 35% higher organic click-through than brands that weren't (0.70% against 0.52%), though Seer says it can't show that the citation caused the difference.

What should you stop paying for?

Stop paying for activity that was a proxy for results. Three line items deserve a hard look at your next renewal.

1. Link-building by volume. If a monthly report counts links acquired, ask what changed in branded search or sales conversations. The correlation data above says reputation beats raw links.
2. Thin content at scale. Dozens of near-identical pages built around keyword variants were the cheap way to rank. They are also the pages an AI summary replaces first, since nothing in them is worth citing.
3. "GEO packages" that promise special tricks. Google says no extra optimization is needed for its AI features. A vendor selling a secret layer on top owes you evidence, and the tools market is covered in our review of generative engine optimization tools.

Not everything is a cut. Technical hygiene, content for questions your buyers really ask, and fixing pages that lost clicks all still earn their keep. If you are choosing an agency, our guide to what the best SEO companies do differently lists the questions to ask.

Keep, cut, start: how to split the budget

There is no honest universal percentage, because it depends on how much of your pipeline comes from search today. This table is a decision aid, not a benchmark.

A workable rule: hold the SEO line flat unless its cost per qualified lead has risen, and fund the new items from the link-building budget rather than from new money.

What should you measure now?

Measure whether buyers know you and choose you, not whether you rank. Rank and raw sessions both lose meaning when clicks per search fall.

• Branded search volume, tracked monthly. It is where AI-influenced research tends to land.
• Click-through on informational pages, not total traffic. Traffic can hold while the share of searches that become clicks drops.
• Citations in AI answers. Ask ChatGPT, Gemini and Perplexity the ten questions your buyers ask and record whether you are named, and next to whom.
• AI referral sessions as their own channel. About 1% of visits today (Conductor), small but separable. Microsoft Clarity found visitors referred by AI assistants signed up for subscriptions at 1.34% against 0.55% for search, though that sample is publisher sites, not B2B (Microsoft Clarity, November 2025).
• Pipeline sourced from organic and from "how did you hear about us." Add a free-text answer to the form. It catches the chatbot mentions that analytics can't.

Which of this can be automated?

The measuring can. Checking what assistants say about your company is the kind of routine a person does once, enthusiastically, and then stops doing in week three.

A weekly run can ask a fixed list of buyer questions to several assistants, log whether and how your company is named, flag a competitor appearing where you used to, and send the changes to whoever owns marketing. This is the shape of marketing workflow automation: you describe the routine in plain language, and we build it, run it and adjust it when the assistants change their answers. It is mostly tested code that asks, stores and compares. AI is used only to read a free-text answer and decide whether your company was recommended or merely mentioned.

If your reporting is already built from analytics exports and spreadsheets, the same logic applies. The Uplift product page shows what that looks like for a team.

Where the evidence runs out

Be careful with anyone who sounds certain, including us. Click-through studies measure informational queries and a handful of datasets. The brand-mention finding is a correlation. The G2 figure is a stated preference from software buyers, and your buyers may behave differently. AI summaries and assistant answers also change faster than annual research can follow.

So test it on your own numbers before you move money. Pull twelve months of Search Console data, split branded from non-branded queries, and see whether your clicks fell where an AI summary appears. If they didn't, you may be fine. If they did, you know which pages to rework first.