SEO in the age of AI: what still works and why

Executive overview

Top-of-funnel search — the long-form content race that defined SEO for two decades — is being absorbed by AI Overviews. Google now answers discovery queries directly, leaving clicks for mid-funnel users who already know what they want.

The durable opportunity is mid- and bottom-funnel SEO: pages that solve a specific problem a buyer has right now, built as a product rather than a content operation.

The core insight: treat SEO as a product decision, not a marketing task — build pages that sit inside the buyer journey, or don't build them at all.

What AI Overviews actually change

• Top-of-funnel queries (best beach vacation, what is a CRM) now get answered in the SERP itself — the click goes away
• Mid-funnel queries retain value: the user already has direction and is ready to act
• E-commerce, local, and transactional pages are far less exposed than informational content
• Sites built on unstructured informational content (WebMD, G2, Capterra) face the largest disruption
• Google delayed full rollout for a year over monetisation, liability, and plagiarism concerns; it is now expanding fast to logged-out and international users

Whether to do SEO at all

• Most B2B SaaS tools should not do SEO — the sales motion is too high-friction for organic conversion
• The right test: can you name what a user would search that leads directly to buying your product?
• If the answer is a blank stare, SEO will not convert — spend the budget elsewhere
• SEO is not free: agency ($10k+/month), in-house headcount, CMS, engineering, and design costs add up fast
• The trade-off question: would the same budget in paid, influencer, or events return money faster?
• SurveyMonkey worked because freemium + search intent + instant sign-up aligned perfectly
• Google Cloud, Mixpanel, high-ACV enterprise tools: wrong motion entirely

The three steps to start

1. Be the user — understand the specific problem they would search before they know your product exists
2. Design the asset — decide whether you need programmatic pages or editorial content based on that user need
3. Build it as a product — SEO pages need design, engineering, and a PM; keyword-stuffed blog posts do not constitute a product

Programmatic vs editorial

• Programmatic combines data sources into pages at scale — right when there is both scale and a clear user use case (Zapier integrations, Tinder city pages, Zillow listings, TripAdvisor properties)
• Editorial is long-form content written per topic — appropriate for media businesses, not most product companies
• Programmatic for its own sake fails: zip-code pages with no demand, template pages that don't connect to what you monetise
• The Zapier lesson: users searched "Gmail + Salesforce connection" — the product already solved that, the SEO just surfaced it
• The Tinder lesson: frame the product around the user's real problem (loneliness in a new city), not the product category (dating app)

AI-generated content

• AI as a writing tool is fine where content was already useful; it makes worthless content cheaper to produce, not better
• Good fit: product descriptions, structured data pages, e-commerce category pages
• Bad fit: generating thousands of blog posts on topics unrelated to the buyer journey
• Google's policy targets unhelpfulness, not AI origin — the quality bar is the same
• Keyword research tools (SEMrush, Ahrefs) can be off by 10x in either direction; use them for relative comparisons and user journey mapping, not absolute forecasts

Forecasting SEO opportunity correctly

• Bottom-up (keyword tool → click-through rate → conversions) is unreliable because the base volume number is often wildly wrong
• Top-down TAM approach: total addressable population → segment by relevance → estimate online purchase rate → target market penetration
• Adjust inputs as you learn; the model stays honest in a way bottoms-up does not
• Useful for board-level conversations and launch decisions in new markets

Appearing in AI Overviews

• Being cited as a brand in an AI Overview is a branding signal, not a direct traffic driver — it means your brand-building is working
• Being linked in an AI Overview means Google sourced your content; click-through is roughly 50/50
• You cannot "optimise" into AI Overviews independently of ranking well in standard results
• Focus on being the best answer for mid-funnel queries; the overview placement follows

Link building and brand

• Buying guest posts on low-authority sites does not work and is inconsistent with believing Google is sophisticated
• Real links come from building something people want to reference — brand and SEO are the same effort
• Google can now match unlinked brand mentions — citations matter as much as HTML links
• PR strategy and SEO strategy should be the same strategy

Google's competitive position

• Google holds 98% of mobile search; Bing spent 20+ years and cannot close the gap
• The DOJ ruling found that default distribution agreements (Apple, Chrome) — not product quality — are the structural moat
• Neva built a better search engine and still failed because of distribution
• LLM startups (OpenAI, Perplexity, Claude) face the same barrier; force of habit and brand keep users on Google
• TikTok and Instagram capture top-of-funnel discovery for Gen Z; Google still owns mid- and bottom-funnel intent

Common myths

• "SEO is free" — it carries real resource costs; evaluate it like any other channel
• "More traffic = more revenue" — traffic that doesn't match your conversion motion is worthless; delete content that pulls the wrong audience
• "Technical SEO fixes algorithm penalties" — most recent Google updates penalise low-quality content; an audit won't fix that
• "Link building is the secret sauce" — bought links on irrelevant sites don't move the needle; brand mentions do
• "PageSpeed is critical" — matters at the margins when competing against fast peers; rarely the real problem
• "Rankings are the KPI" — rankings are a proxy; the actual metric is whatever business outcome SEO is supposed to drive

Getting started without a background in SEO

• PMs without SEO expertise have driven large results (Airbnb example) by applying product thinking to search
• Hire a growth advisor to compress the learning curve, or hire a junior in-house and expect slower progress
• Agencies are easiest to evaluate on deliverables but often default to content volume — which may not be the right deliverable
• Build a product roadmap with milestones; missed milestones are the real signal that SEO is stalling, not rankings
• SEO is not dying — user-directed information retrieval is structural; what's dying is the top-of-funnel content farm model