GEO vs SEO: what actually changes (and what doesn't)

The honest version: most of your SEO work still counts. Crawlability, clean semantic HTML, structured data, internal linking, topical depth, actual expertise, a fast site. All of that feeds directly into whether an AI engine can find and use your content. If you've done the SEO fundamentals well, you are not starting from zero on GEO. You're starting from maybe 60%.

The dishonest version, which you'll hear a lot, is that SEO is dead and GEO replaces it. That's wrong, and it leads people to abandon the exact infrastructure their AI visibility depends on.

What's actually true is narrower and more useful: the ranking still works, but the ranking stopped paying.

What is the core difference between GEO and SEO?

SEO competes for a slot on a page of links. GEO competes for a sentence inside a generated answer.

That sounds like a semantic distinction until you follow it through. On a results page, ten brands each get a slot, and the user chooses. In a generated answer, the model has already chosen. It names three or four brands and describes them. There's no scroll, no page two, no long tail of visibility for whoever came eleventh. You're in the answer or you don't exist for that query.

The last row of that table is the one people underestimate. Answers are probabilistic. Ask the same engine the same question three times and you can get three slightly different brand lists. This means a single check tells you nothing, and it means "I asked ChatGPT and we came up first" is not a measurement. It's an anecdote.

What carries over from SEO unchanged?

Roughly this list, and it's a long one:

• Crawler access. AI engines run their own crawlers (GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended). If those are disallowed in robots.txt, everything downstream is moot. Check this first, because a lot of sites blocked them defensively in 2024 and never revisited it.
• Server-rendered content. AI crawlers are less patient with client-side JavaScript than Googlebot is. If your key content only exists after hydration, assume it's invisible.
• Structured data. Organization, Product, FAQPage, Article, BreadcrumbList. Schema doesn't magically get you cited, but it gives the machine an unambiguous statement of what you are, and consistency across schema, page copy and third-party listings is what builds a stable entity.
• Topical authority. Engines still lean on domains that demonstrably cover a subject deeply. One good post about GEO on a site otherwise about dog food is not going to get retrieved.
• Genuine first-hand expertise. The E-E-A-T stuff. Original data, named authors, real experience. This got more valuable, not less, because it's the thing that makes a passage worth lifting.
• Site speed and clean HTML. Same reasons as always.

If someone tells you GEO means throwing out your technical SEO, they're selling something.

What is genuinely new?

Three things. Everything else is a variation on them.

1. Consensus beats a single great page

In SEO you can win position one with one exceptional page. In GEO, one exceptional page that no one else corroborates is a single unsupported source, and models are cautious about naming a brand on the strength of the brand's own marketing.

What moves an AI answer is agreement across independent sources. When four listicles, two Reddit threads, a review platform and a news mention all describe you the same way, the model has enough corroboration to state it as fact. When only your own site makes the claim, it usually doesn't.

The practical consequence: a meaningful share of GEO work happens off your domain. Getting added to the "best X tools" roundups that engines already cite is often higher leverage than publishing another post. This is uncomfortable for SEO teams who are used to controlling their own surface, but it's where the wins are.

2. The extractable passage replaces the ranked page

Models don't cite pages. They lift passages. A 2,000-word article gets chunked, and the chunks compete individually.

This changes how you write. The first one or two sentences under every heading have to answer that heading, directly, before any context. If the answer to "how much does it cost" arrives in paragraph four after three paragraphs of philosophy about value, the chunk containing the heading doesn't contain the answer, and the chunk containing the answer doesn't contain the question. Neither one is liftable.

Concretely, do this:

How much does Spottlo cost?

$39/month base, plus $19/month per additional brand. That includes 25 tracked prompts per brand across ChatGPT, Perplexity, Gemini and Google AI Overviews, with weekly scans.

Not this:

Pricing

We believe pricing should be simple. Many tools in this space have complicated tiers that make it hard to know what you're paying for. That's why we thought carefully about our approach...

Same information, one of them is citable.

3. Measurement is sampling, not tracking

Rank tracking is deterministic. You check position, you get a number. AI answers are generated, so you have to sample: run a fixed prompt set, on a schedule, across each engine, and look at rates rather than events.

The metrics that replace "average position":

• Mention rate — of your N tracked prompts, what percentage produced an answer naming you?
• Share of voice — of all brand mentions across your prompt set, what percentage are you? This is the competitive number, and it's the one executives understand.
• Position in answer — named first, or named fourth in a list of six?
• Sentiment — recommended, mentioned neutrally, or mentioned as the thing to avoid.
• Citation share — how often your domain is one of the linked sources, which is distinct from being mentioned by name.

You can do this by hand for a week. You can't do it by hand for a quarter across four engines, which is the entire reason tools like Spottlo exist. If you want the fast version, the free report runs your prompts once so you can see the shape of the problem before deciding whether it's worth automating.

What SEO tactics should I actually stop doing?

A short list, but a real one:

Is the traffic loss actually a loss?

Partly, and the compensation is real. AI answers send you less traffic but better traffic, because the model has already done the qualification step that used to happen across five open tabs.

Semrush found an AI-search visitor is 4.4x as valuable as an organic search visitor by conversion rate. Adobe measured AI-referred traffic to US retail sites up 138% year over year and converting 54% better than non-AI traffic. And when OpenAI started putting clickable brand links in ChatGPT answers, Similarweb saw referral traffic rise 157.7% week over week, which suggests the click volume is partly a product decision by the engines rather than a permanent physical law.

So the picture isn't "traffic is gone." It's "volume down, quality up, and the entry ticket is being named in the answer."

You'll also see the Gartner line quoted a lot: that search engine volume would drop 25% by 2026. Worth being precise here. That was a 2024 prediction, not a measured outcome, and it gets cited as if it were data. Don't build a strategy deck on it.

How much of my effort should go to each?

If you're starting from a solid SEO base, a reasonable split for the next few quarters is 70% keep doing what you're doing (technical health, depth, expertise) and 30% net-new GEO work. That 30% is roughly: build and run a prompt set, audit the sources engines cite for your category, get into the listicles and review platforms that show up in those citations, and rewrite your highest-intent pages answer-first.

Note what's not in that 30%: a rebrand of your content calendar, a new team, or a llms.txt file. Publish llms.txt if you like, it costs nothing, but don't confuse it with a growth lever.

What to do next

1. Audit robots.txt for AI crawlers today. GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended. This is a five-minute check that gates everything else.
2. Rewrite your top five commercial pages answer-first. First sentence under every heading answers the heading. Nothing else changes.
3. Build a 25-prompt buyer question set and baseline it across all four engines so you know your real mention rate rather than your guessed one.
4. List the sources cited in every answer where a competitor beat you. Those URLs are the actual work. Go get included in them.
5. Add share of voice to whatever dashboard already shows organic position, so the two scoreboards sit side by side. If you're evaluating tools for it, compare them on engine coverage first, because most of the category charges more per engine.