Is SEO dead? What AI answers changed and what they did not

No. The question "is SEO dead" keeps getting asked because the answer box replaced the blue link, not because ranking stopped mattering. Google handles over 15 billion searches a day, and AI answer engines build their replies from pages they find in search. In a test of 434 live answers, engines cited 3.4 to 10.1 sources each. What changed is who gets the click.

Is SEO dead or just changing?

Start with scale, because scale settles most of this argument.

Google controlled 89% of all U.S. web traffic in 2025. Search volume sits somewhere over 15 billion queries a day. AI and LLM searches account for about 6% of global search volume right now.

That 6% is growing fast. It's about triple what it was a year ago. Triple again and it's still a minority of where people look things up. So the honest read is: a large, slow channel is losing some share to a small, fast one, and both of them run on indexed web pages.

Meanwhile Google is shipping 12-plus algorithm changes per day. A dead channel doesn't get changed twelve times a day.

The thing people actually mean when they ask the question is narrower: "will my informational blog post still bring me visitors?" For a lot of posts, no. That's a different sentence from "SEO is dead."

What AI Overviews and AI Mode actually changed

Google rolled out AI Overviews across its core search experience, and at Google I/O it announced a mass launch of updates built on AI summarized results. The consequence is mechanical: a summary sits above the results, so the user gets their answer without a visit.

Zero-click searches are now the norm for informational queries. And top-of-funnel informational content has lost most of its commercial value.

Read that carefully. It says the traffic from "what is X" posts stopped converting. It does not say the pages stopped being read. They get read inside the answer box, by the summarizer, and sometimes by the buyer who sees your name in a list of three.

So the loss is real and it is specific: the cheap top-of-funnel article as a traffic engine. If that was the whole strategy, the strategy is in trouble. If it was one layer, the layer changed shape.

What hasn't changed at all

Answer engines do not invent vendor lists out of nothing. They fetch pages, read them, and cite them.

In welaunch's test of 434 live AI answers to buyer questions across seven software categories, run 27 Sep 2026 in the US, ChatGPT cited 5.8 sources per answer on average over 110 answers. Perplexity cited 10.1 per answer over 103 answers and gave no source in zero of them. Gemini averaged 9.2 over 110 answers. Google AI Mode was the leanest at 3.4 over 111 answers.

Every one of those citations is a URL that a crawler reached, parsed and stored. Which means the boring parts of SEO still decide whether you are eligible to be mentioned at all. Pages that render server-side. Pages that aren't blocked. Pages with a clear title and a direct first paragraph. Pages that exist on a domain a crawler has reason to visit.

If a page can't be crawled, it can't be cited. That rule did not change.

Is SEO dead with AI? What citation data from four engines shows

People asking "is SEO dead with AI" usually imagine the model answering from memory. Sometimes it does. Mostly it retrieves.

Here's the per-engine picture from the same test of 434 answers, 27 Sep 2026, US.

Source: welaunch test of live AI answers, 27 Sep 2026.

Two things fall out of this table.

First, the sourceless answer is rare. Perplexity never gave one in 103 answers. Google AI Mode gave 21 out of 111, the highest of the four, and it also cites the fewest sources when it does cite. Fewer slots means harder competition for each one.

Second, the spread between 3.4 and 10.1 is the difference between a shortlist and a sweep.

Is GEO replacing SEO, or sitting on top of it?

Sitting on top. Generative engine optimization, answer engine optimization, whatever the label, it depends on the same inputs.

The engines cite pages. The pages come from an index. The index is built by crawlers that follow the same rules they followed before AI Overviews shipped across core search. A GEO program with no technical SEO underneath it is a program with no supply of citable pages.

What GEO adds is a different target. Classic SEO aimed at a position in a list of links. GEO aims at being named inside a paragraph the user reads, and at being the source the model picked from the 3.4 to 10.1 it had available.

Those are not the same target, and they do not always come from the same page. Which brings us to where the citations actually come from.

Where AI engines actually get their answers from

A lot of them come from places you don't own.

Across all 434 answers in welaunch's test, 106 (24%) cited reddit.com. Per engine, that was 40% of Google AI Mode answers, 31% of Perplexity, 20% of Gemini and 6% of ChatGPT.

Narrow it to buying questions and the pattern sharpens. Of 80 answers to questions about choosing an AI visibility tool, 22 cited reddit.com. Of 63 answers to AI-visibility buyer questions, 12 cited reddit.com and 12 cited youtube.com.

So a forum thread and a video can both outrank your product page as a source, because they read like someone comparing options rather than someone selling one. That's not a trick to copy. It's a signal about what the retrieval step is looking for: a page where your product appears in context, next to alternatives, described by someone who isn't you.

Directories and roundups do the same job. So do comparison pages on third-party sites.

Why the same brand doesn't show up in the same answer twice

This one surprises people who check their visibility once and draw a conclusion.

In welaunch's test, each prompt was run four times on each engine. Only 9 of 32 brand appearances in ChatGPT answers (28%) were named in all four runs of the same prompt. On Perplexity it was 6 of 21 (29%).

Measured as overlap between the brand sets returned, ChatGPT's four runs of the same prompt matched with a mean Jaccard similarity of 0.54 across the 7 prompts that named any brand. Perplexity came in at 0.57 across 5 prompts.

Roughly half the names change between runs of an identical question.

Practically: one check tells you almost nothing. If you ask once and see your competitor, that may be noise. If you ask once and see yourself, that may also be noise. You need repeated sampling of the same questions over time before any of it means something, which is the whole reason tools in this space run prompts on a schedule rather than on demand.

Does SEO still matter in 2026? What brand signals are doing now

Yes, and the weighting moved. Brand mentions, reviews and direct traffic now correlate more strongly with rankings than they did three years ago.

That lines up with what the citation data shows. The pages engines pull from are often the pages where your brand is being discussed rather than described. A Reddit thread, a review listing, a comparison post. Those are brand mentions by another name.

So the work converges. Getting mentioned in the places buyers read helps rankings, and the same mentions are what the answer engines retrieve. One effort, two outputs.

What real traffic data from one brand shows

Numbers from a single brand, RefiJet, published by Neil Patel, show the channels moving together rather than trading off.

SERP features up 30,800% since May 2024. Rankings in positions 1 to 3 up 522% year over year. Traffic from LLMs up 2012% year over year, with site-wide page views from LLMs up 7144%.

And the part that pays rent: funded loans from organic search and LLMs up 178% year over year.

One brand is one brand. But the shape is worth noting. The LLM traffic grew alongside the classic ranking gains, not instead of them. Whatever earned the position 1 to 3 slots also made the pages retrievable.

For context on why positions still matter at all, SmartInsights data cited in the same piece puts double-digit click-through rates on the top 3 positions.

What to measure instead of rankings alone

Rankings still belong on the dashboard. They just stopped being the whole dashboard.

Track share of answers, not share of results. In welaunch's test, Profound was named in 23 of 63 answers to AI-visibility buyer questions, with Otterly and Peec at 17 each. That's a measurable position in a market, derived entirely from counting named brands across repeated runs.

Track it per engine. A tool that shows up constantly on Perplexity, which cites 10.1 sources per answer, may be invisible on Google AI Mode, which cites 3.4.

Track which sources get cited for your questions, separately from whether you're named. The citation list is your target list. If Reddit shows up in 40% of Google AI Mode answers for your category, you now know where the work is.

And track branded search and direct traffic, because those signals correlate with rankings more than they used to, and because a zero-click answer that names you often shows up as someone typing your name later.

What I'd do if SEO were my job starting today

Monday morning, in order:

1. Write down the ten questions your buyers actually type before they buy. Not keywords. Questions.
2. Ask each one on ChatGPT, Perplexity, Gemini and Google AI Mode. Four times each, because roughly 28% to 29% of brand appearances held across all four runs in welaunch's test of live AI answers, 27 Sep 2026. Record every source cited, not just whether you were named.
3. Sort the sources by how often they appear. That's your list.
4. Fix crawlability on anything you own that appears near the top of that list, and on your comparison and pricing pages.
5. Get onto the third-party pages that keep appearing. Listings, roundups, forum threads where the question is already being asked.
6. Stop writing top-of-funnel definitional posts as a traffic play. They've lost most of their commercial value.

Steps 2 through 5 are a weekly chore if you do them by hand. welaunch does them on a schedule: it asks your buyer questions on all four engines several times a week, keeps every answer in full so you can read what each engine said, finds the pages they cite, and drafts the reply or listing that gets you onto one. It works from your accounts, so you review every draft and press submit yourself.

Where to get listed so AI engines can cite you

Directories are an unglamorous but direct route onto cited pages. They rank, they get crawled, and they list you next to alternatives, which is the format retrieval seems to like.

welaunch's directory database has 273 launch and listing directories. Of those, 198 accept a free listing, 16 are free only with a badge or backlink, and 56 charge.

By category: 163 SaaS and startup directories, 95 AI tool directories, and 15 agent and developer directories. 100 of the 273 accept AI products only, which matters if you aren't one.

On forums it's a different job. welaunch doesn't stuff links into Reddit or Quora replies. On those platforms it names the product only, with disclosure, because a thread that reads like an ad gets removed and then cites nobody.

FAQ

Is SEO a thing anymore?

Yes. Google still handles over 15 billion searches a day and controlled 89% of U.S. web traffic in 2025. AI search is about 6% of global search volume. Both channels pull from the same crawled index.

Is local SEO dead?

Nothing in the data here covers local specifically. What does apply: AI answers are assembled from crawlable pages and citations, so listings, reviews and crawlable location pages feed both ranked results and generated answers. Brand mentions and reviews correlate more strongly with rankings than three years ago.

How long will SEO exist?

As long as answers are built from retrieved web pages. Across 434 live AI answers tested in September 2026, engines averaged between 3.4 and 10.1 cited sources each, and Perplexity gave a sourceless answer in none of its 103. Retrieval means somebody is optimizing for retrieval.

Will SEO exist in 5 years?

The targeting will keep shifting from link positions toward being the named source inside an answer. Given Google ships 12-plus algorithm changes per day, expect the tactics to turn over several times. The underlying job, making pages a machine will fetch and quote, doesn't look like it's going away.

What role do brand signals play in modern SEO?

A bigger one than before. Brand mentions, reviews and direct traffic now correlate more strongly with rankings than they did three years ago. They also overlap with what AI engines retrieve: 24% of the 434 answers tested cited Reddit, which is mostly brand discussion rather than brand publishing.

Why is search everywhere optimization becoming important?

Because the four engines behave differently. Google AI Mode cited reddit.com in 40% of its answers while ChatGPT did so in 6%, and source counts per answer ranged from 3.4 to 10.1. Optimizing for one engine's habits leaves you missing from the others.

Sources

• Is SEO Dead in 2026? (A Data-Driven Answer) | Neil Patel
• Is SEO Dead? Here's What I'm Telling Every Client Who Asks Me in 2026
• SEO Is Dead, According to Google
• welaunch.sh
• welaunch test of live AI answers, 27 Sep 2026
• welaunch's test of 434 live AI answers to buyer questions in seven software categories, 27 Sep 2026, US

Related on welaunch

• SaaS marketing in 2026: the channels that still work for a small team
• SaaS content marketing: the content types to write, in order, to get ranked and cited
• AEO vs GEO: the same job, two names, and where they actually differ