Start From SEO, Not From Zero

Every AI SEO program proposal lands on an executive desk carrying the same implication: this is new, so it needs new budget, new tools, and maybe a new team. Before approving any of that, settle the first design decision — how much of this is actually new for your organization?

What your SEO already bought

Google’s own guidance is blunt about the foundation: optimizing for generative AI features in Search “is still SEO.” The AI answers users see are grounded in the same retrieval systems the SEO team has been optimizing for years. ChatGPT search discovers pages through Bing’s index. Google’s AI Overviews and AI Mode draw on Google’s index and ranking systems. A page that cannot be crawled, rendered, and indexed cannot be cited — by anything.

That means a site with competent traditional SEO already owns most of the load-bearing assets: a crawlable estate, indexed content, a structured-data habit, an editorial quality bar, and the discipline of reading Search Console before making claims. None of that gets rebuilt. All of it gets inherited.

This is why a surprising share of “AI visibility problems” dissolve under ordinary SEO diagnosis — a pattern covered in Why Many AI Visibility Failures Are Just SEO Failures.

What is genuinely new

Five things do not carry over, and they define the program’s real scope:

• The unit of competition changed. Traditional SEO competes for a ranked position. AI surfaces compete for retrieval, extraction, and citation inside a synthesized answer. A page can rank well and still lose the citation because its answer is buried, hedged, or split across boilerplate.
• The surface count multiplied, and the surfaces disagree. ChatGPT, Google’s AI features, Copilot, Perplexity, and marketplace assistants ground on different indexes and cite different sources. Semrush’s reasoning-mode study found only 25.6% of cited domains overlapped between ChatGPT’s two reasoning modes — for identical prompts, on one product. Optimizing “for AI” as a single target is a category error.
• Success decoupled from traffic. Pew Research measured Google users clicking a result in 8% of visits when an AI summary appeared, versus 15% without one — and clicking a link inside the summary in just 1% of visits. A brand can win the answer and see nothing in analytics.
• The model layer exists. Part of what an assistant says about your brand comes from training data, not live retrieval. Page edits cannot reach it; only sustained entity and earned-media signals migrate into the next training run.
• Measurement has no Search Console. Google reports AI Overview and AI Mode traffic inside overall Search totals, not as a separate line. First-party citation data exists only on Microsoft’s stack. Everything else is sampled by third parties.

The decision

Size the program as a delta on top of SEO, not a parallel discipline. In practice:

• If your traditional SEO is genuinely competent, the new spend concentrates on the five deltas above: answer-level content structure, per-surface strategy, measurement design, entity work, and governance.
• If it is not, fix that first. Every AI surface inherits your crawl, index, and quality problems at full price, and no AI-specific tactic compensates for them.
• Reject any proposal priced as a rebuild. A vendor or internal champion who cannot articulate what your current SEO already covers has not scoped the work — they have renamed it.

The rest of this section walks the decisions the delta actually requires: which surfaces deserve budget, who owns the work, what to build first, how to measure it, what it costs, and what can go wrong. Then the project menu turns those decisions into a chartered selection — and the myths guide lists what to strike before anyone charters it.