Will AI Replace SEO? What Our Traffic Data Actually Shows

No. AI search isn't replacing SEO. It's replacing the version of SEO that chased rankings instead of answers.

We've tracked both on mintec.co for six months, and the split is nowhere near as dramatic as the headlines. Google still sends us about 130 clicks a week. ChatGPT, Gemini, Claude and Perplexity send 16 sessions. Roughly eight to one, in favor of the old way.

The interesting part isn't the ratio, though. It's the direction. The organic number wobbles with every Google update. The AI number only grows, and the people who arrive from it behave differently. So when a client asks whether we should keep doing SEO, our answer is: keep doing it, but stop doing half of what you're doing now.

The question that shows up in every sales call

Lately every first call with a prospect includes some version of "we heard AI search is killing SEO". Usually they've seen a headline, or a LinkedIn post with a screenshot of ChatGPT answering a question without citing anyone.

The fear is reasonable. The framing is wrong. AI search isn't taking traffic from sites like ours. It's rerouting the traffic that used to go to sites that didn't deserve it. If a page's only job was to sit at position 5 and collect clicks from people who already knew what they wanted, that page has a problem now, and it should.

What six months of our own numbers show

We run a GEO and SEO program on our own site, which means we see both sides of the ledger at once. The weekly report looks like this:

• About 130 clicks a week from Google organic, most of it from posts that answer a specific question.
• 16 sessions a week that arrive straight from AI answers. ChatGPT sends 11 of them, then Gemini, then Claude. Perplexity hasn't sent a single session in months, even though it cites us.
• The pages AI cites most are not our blog posts. They're service pages and contact pages with direct answers. Our how-to-get-cited breakdown shows the pattern in detail.
• We audited all 501 published posts for citability and flagged 181 as high-citability. The posts AI engines actually cite share a format: a direct answer up top, a FAQ block, a comparison or table, and links to sources. When we ran a test with seven prompts, the same two posts got cited every single time. Format did more work than topic.

One example we keep using internally: our Spanish post about digital marketing costs in Latin America is the single most-visited page from ChatGPT, and the second one is our contact page. Nobody links to those pages. The models just found them because they answer a question cleanly.

Rankings and citations have come apart

The most confusing thing we see in client accounts is a page ranking first in Google while AI engines never cite it, and the reverse: a page on page 3 getting quoted by ChatGPT every week. That decoupling is now the rule, not the exception.

This is why "check your positions" is dying as a KPI. Position tracking tells you what Google thinks of you. It tells you nothing about what a model thinks of you, and the two are barely correlated anymore. We wrote about which metrics actually hold up here, and the short version is: citations, AI referrals, and what those referrals do on your site.

There's also a trust layer now. Only 28% of Americans trust AI search, according to a Search Engine Journal survey, which means a citation is only the beginning. The user still opens the cited page to verify. Pages that survive that verification click keep the traffic. Pages that don't lose it to the model itself.

The parts of SEO that are dying (and the parts that aren't)

If I had to split everything we know into three buckets:

Dying. Keyword-stuffed pages written for crawlers. Position reports as the goal of the month. Link schemes that buy authority nobody verifies. Meta descriptions written for search results nobody reads anymore.

Surviving. The technical base: crawlability, speed, schema. Trust signals: real authors, publish dates, named sources. Content that answers the actual question a person typed. Internal linking that helps crawlers and models alike understand what a page is about.

New. Self-contained answer blocks with FAQ and tables. A llms.txt file if you want control over how models index you. Separate tracking for AI referrals so you can see what's working. And a refresh cadence, because AI engines stop citing content that goes stale. We measured 60-day citation decay on our own posts, which is faster than most people expect.

Google's own guidance keeps saying GEO is still SEO, and we mostly agree. But the official line was written to keep you calm. The part worth taking seriously is that answer surfaces, not blue links, are where attention is moving, and Google's Gemini-era updates keep pushing in that direction.

Even Google feels the pressure

The "AI will replace Google" story is usually told from the user side. The business side is just as real. eMarketer projects Meta will pass Google in digital ad revenue this year, and Google's response is to push AI Overviews and answer surfaces harder, which eats into the clickable results that used to be guaranteed.

And the floor can fall out without warning. In late July, a government-backed tourism site lost 73% of its Google traffic in 48 hours. No manual action, no announced update, all technical checks clean. Googlebot still crawled the site normally. Bing and ChatGPT traffic stayed flat. The site did nothing wrong and lost three quarters of its organic visits anyway.

We can't explain that one, and neither can the SEO community. What it does tell you is that nobody's Google traffic is guaranteed. AI visibility is starting to look like the hedge.

What we tell clients today

Four steps, in order:

1. Answer one question per page, in a self-contained block that works even if nobody clicks. It's the biggest citability lever we've found.
2. Keep the technical base clean. Schema, speed, crawlability. Models inherited the same respect for structure that Google always had.
3. Track AI referrals separately. If you can't see which pages the models cite and send traffic to, you're optimizing blind.
4. Refresh on a cadence. Citation decay is real and it's fast. Sixty days, in our data.

The verdict

Will AI replace SEO? No. It's replacing position-chasing. The sites that lose are the ones that optimized for the search results page instead of for the answer itself.

Our own numbers say the old channel still works, and the new one is small but compounding. The strategy that survives both is boring: answer questions better than the models can, stay technically clean, and measure both channels separately.

That's the whole job. It always kind of was.