AI-Powered Content Strategy: From SEO to GEO

Overview

Search is fundamentally broken, and AI engines are rewriting the rules of visibility. We are looking at Isabelle Guis's Stanford Continuing Studies Fall 2026 lecture, Week 3. Her deck, titled 'AI-Powered Content Strategy: From SEO to GEO', breaks down exactly how to adjust your marketing activities. The old playbook of

In this show

• Search is fundamentally broken, and AI engines are rewriting the rules of visibility. We are looking at Isabelle Guis's Stanford Continuing Studies Fall 2026 lecture, Week 3. Her deck, titled 'AI-Powered Content Strategy: From SEO to GEO', breaks down exactly how to adjust your marketing activities. The old playbook of keyword stuffing and backlink farming is dead. To survive Answer Engine Optimization, you have to understand what these new platforms actually value. I'm joined by two digital strategy experts to dissect every single slide of this presentation.
• This title slide sets the core agenda for Week 3: 'What activities each platform values.' Guis is shifting the focus away from generic internet traffic. If you want your brand to show up as a definitive answer in Claude or ChatGPT, you have to reverse-engineer their specific ingestion methods. We are moving from a world where Google's algorithm dictated everything, to a fragmented landscape of distinct AI models with entirely different appetites for data.
• Look at this matrix showing Data Ingestion Methods as of October 2026. Guis breaks down five platforms: ChatGPT, Perplexity, Gemini, Claude, and Copilot. Notice how varied their sources are. Wikipedia is heavily weighted across the board, either as training data or live retrieval. But look at YouTube: Gemini has native, default access since October 2025, while Claude has no native access. If your primary asset is a video, Gemini will synthesize it beautifully, but Claude won't even see it.
• Here is the master heatmap of traditional marketing methods against AI source types. Green is high leverage, red is none. Notice the top row: Earned PR, meaning genuine press coverage, is 'High' for public websites and news sites, and 'Med-high' for Wikipedia and Wikidata. Conversely, paid advertising at the bottom is completely red. No AI platform trains on or retrieves your paid ads. The models are optimized to filter out paid placement and surface authoritative, earned information.
• This slide digs into Public Websites, Retrieval, and Freshness. Guis outlines the exact retrieval mechanisms. Perplexity always uses its own crawler with a strong freshness bias—near real-time. Copilot grounds everything in the Bing index by default. Meanwhile, ChatGPT's behavior is split; it has a frozen training snapshot, now GPT-5.6 by default, but uses live Bing for browsing. Understanding these cutoff dates and live-fetch triggers dictates whether a product launch actually registers in an AI's answer.
• Guis returns to the heatmap, circling Earned PR and SEO. Earned PR is the highest-leverage method overall. But look at SEO and technical optimization: it ranks 'High' for public websites, but 'None' for Wikipedia, News, or YouTube. Technical SEO still matters, but it strictly controls crawlability for the bots, not your actual domain authority. It gets the AI to read your page, but it doesn't convince the AI that your page is the ground truth.
• This slide contrasts Classic SEO Logic against New GEO Logic for Earned PR. The old goal was reach, measured in impressions and backlinks for domain authority. The new GEO logic treats every placement as a potential Wikipedia or Wikidata source. The metric is now citation frequency and share of voice across AI platforms. Instead of pitching narrative brand storytelling, marketers must push for specific, quotable factual claims—numbers and named comparisons that an LLM can easily extract.
• Looking at SEO/Technical Optimization as of September 2026, Guis lists the specific bots you must allow. ChatGPT uses OAI-SearchBot. Claude uses ClaudeBot tied to Brave's index. Gemini uses Google-Extended. The yellow box at the bottom is critical: since mid-2026, Cloudflare blocks Training and Agent bot categories by default, while keeping Search bots allowed. Managing your robots.txt and infrastructure settings is now a prerequisite. If you block these specific fetchers, you simply do not exist in the AI's live retrieval.
• Back to the heatmap, we are now focusing on 'Company blog / owned content' and 'Original research / data studies.' Both show up as 'Med-high' for public websites. Original research is particularly strong because it pairs well with academic publishing, and as noted here, Reddit is now a second distribution path for those data studies. Your owned content isn't building training trust, but it is highly retrievable if structured correctly.
• This slide explains how to restructure that Company Blog. Classic SEO demanded keyword-optimized H2s and word-count targets to lower bounce rates. The New GEO Logic demands answer-first paragraphs. You need one discrete claim per sentence, using definitional statements like 'X is...'. The success signal is whether your exact wording survives into an AI answer. Being first and consistent is unusually high-leverage here, because wording consistency matters more to an LLM than traditional backlinks.
• Guis provides brilliant implementation examples here. For a definition statement, the classic SEO logic is fluffy: 'In today's fast-evolving digital landscape...' The new GEO logic is surgical: 'Cloud-based inventory management software is a system that tracks stock levels in real time over the internet.' It defines the entity immediately. For positioning, instead of a full narrative blog, the GEO approach uses a dense claim: 'Unlike Competitor, Product supports real-time multi-warehouse sync without requiring a third-party integration.'
• More implementation examples on this slide. Look at the Data Points. Classic SEO writes three paragraphs summarizing a survey. The GEO logic compresses it into a highly citable sentence: 'In a March 2026 survey of 500 US retailers, 43% reported inventory discrepancies costing over $10,000 annually.' It includes the date, sample size, and brand name. For FAQs, instead of generic questions, use direct formats: 'How much does implementation typically take for a mid-size retailer?' AI engines lift these exact structures.
• Now we enter the Industry Examples using Claude, comparing Salesforce and HubSpot. Look at how Claude summarizes Salesforce: it's a dry timeline of acquisitions, noting it was founded in 1999 by Marc Benioff, and lists Service Cloud and Agentforce. The red box highlights a classroom discussion takeaway: entity recognition isn't equal. AI engines synthesize brand identities using fundamentally different primary source types. Salesforce's positioning is driven by earned PR and financial history.
• This slide reveals exactly where Claude pulled that Salesforce summary from. The sources are Wikipedia, the SEC Form 10-K, and financial-data aggregators like MacroTrends and Stock Analysis. Claude cited absolutely no major news publications or traditional newspapers for this corporate query. If your corporate communications team is spending all their time pitching product features to tech blogs, they are missing the fact that the AI is learning about you from your SEC filings and Wikipedia page.
• Contrast that with the HubSpot example on this slide. HubSpot's summary mentions their AI layer, Breeze, and specific products. Their sources include HubSpot corporate content, new publications, and Business Wire press releases. The slide notes that despite a huge backlash over a price increase attempt in July 2026, HubSpot still owned their narrative. They successfully fed the AI their own press releases and community content to shape the immediate retrieval results.
• This slide shows a direct Claude query: 'Why would someone choose Hubspot over Salesforce?' Notice the output. It uses bolded, thematic headers like 'Ease of use and faster time-to-value' and 'Included, simpler AI.' As Guis points out, these are built straight from HubSpot's own positioning content. It is not a neutral summary. HubSpot seeded these specific comparative claims on their own properties, and the model simply lifted and paraphrased them as objective advice.
• Now look at the inverse query on this slide: 'Why would someone choose Salesforce over HubSpot?' The response is structured the way outside reviewers frame it, focusing on 'deep customization' and 'complex data governance.' Why? Because Salesforce isn't the one writing this comparison. The critical takeaway here is in bold: If you don't author the comparison, the model still writes one. It just writes it from whoever else has, leaving your brand vulnerable.
• Guis breaks down the PR contrast in this table. For Earned PR, HubSpot generated independent analyst coverage about their AI strategy, like a blog post titled 'Is HubSpot Quietly Winning the AI Race?'. Salesforce's earned coverage is heavier and more frequent, primarily driven by CNBC and the financial press, but it's framed entirely around earnings and stock price, not product commentary. HubSpot is actively shaping the AI-strategy narrative, while Salesforce is riding financial momentum.
• This slide shows the exact asset HubSpot benefited from: a July 2026 blog post by Hypha titled 'Is HubSpot Quietly Winning the AI Race?' This is an independent third-party site writing deep, contextual analysis about HubSpot's specific AI tools like Breeze and AEO. When Perplexity or Claude looks for qualitative assessments of HubSpot's product momentum, this is the exact type of high-signal, text-heavy content the models prioritize for retrieval.
• And here is the corresponding Salesforce asset. It's a May 2026 CNBC article: 'Salesforce beats on earnings and revenue but full-year guidance comes in light.' It's purely financial. While it mentions the Agentforce annualized revenue passing a billion dollars, the entire context is stock performance. AI engines index this rapidly for real-time news queries, but it does very little to train the model on the nuance of how Salesforce's CRM actually works for end-users.
• Moving to the next criterion: SEO/GEO technical optimization. HubSpot built a standalone AEO product page designed purely as citation-bait. Salesforce, on the other hand, relies on a 'What Is CRM?' glossary page. The contrast is stark. Salesforce has a massive, solid definitional asset, but they are optimizing for a broad category term that they have to share credit for with every competitor. HubSpot is optimizing for a brand-new term they can own outright.
• Here is the screenshot of that HubSpot page. 'HubSpot AEO - Answer Engine Optimization Software.' By planting their flag on this specific acronym and concept, they are executing a 'first mover on a new concept' strategy. When an LLM is prompted to define AEO, there is almost zero competing literature. HubSpot becomes the definitive, uncontested ground truth for the definition, virtually guaranteeing they are cited as the primary source in AI summaries.
• This is the Salesforce page for comparison: 'What Is CRM (Customer Relationship Management)?' It's a classic, high-traffic SEO play. It defines the system cleanly. But from a generative AI perspective, the model has ingested ten thousand different pages defining CRM. Salesforce's page is just one drop in that massive training bucket. It doesn't give them a unique semantic edge in an AI chat interface the way HubSpot's hyper-specific AEO page does.
• This table covers Blog/editorial and Research. HubSpot has evergreen definitional glossary posts, like 'What is inbound marketing?'. Salesforce has no close equivalent; their public blog leans toward product updates and customer stories. However, on Research, it's a tie. HubSpot's State of Marketing report and Salesforce's State of Sales, 7th edition, are comparable in rigor and cadence. This is the one row where Salesforce is genuinely on par, producing data-heavy assets AI engines love.
• Looking at the HubSpot asset on this slide, 'What is inbound marketing?', you can see the execution. HubSpot literally coined this term years ago. They own it as a definitional page. The text is clean, structured, and reads like an encyclopedia entry. Salesforce simply doesn't have an equivalent single term that it owns in the same way in its public content, which puts them at a disadvantage when LLMs are mapping semantic entities.
• This slide shows HubSpot's '2026 State of Marketing Report.' Original research is gold for Answer Engine Optimization. LLMs crave statistics, percentages, and firm geographic or temporal data points. By publishing a massive, data-dense report, HubSpot creates a reference node. Every time an AI needs a stat on AI adoption in marketing, it pulls from this report, citing HubSpot as the authority and driving high-intent brand visibility.
• And here is Salesforce's equivalent, the 'State of Sales Report.' Notice the bullet points: 'Why 94% of sales leaders say agents are essential to growth.' This is exactly what Guis was talking about earlier regarding discrete, surgical data points. Salesforce is feeding the AI clear, unambiguous statistical claims. Even though it's behind a lead-capture form, the abstract and landing page copy give the AI crawlers exactly what they need to ground a response.
• Guis brings us back to the heatmap one last time, circling Wikipedia engagement. Notice it is rated 'High' for public websites and 'Med-high' for the Knowledge Graph, but completely red for everything else. This highlights a critical pivot. Wikipedia is no longer just a nice-to-have profile; it is the foundational layer of truth for models like Gemini's Knowledge Graph and ChatGPT's baseline entity understanding.
• This slide contrasts Wikipedia strategies. Classic SEO treated Wikipedia as a backlink to acquire, often using paid agencies to heavily edit pages. New GEO logic treats Wikipedia as a reputation to earn. You must be the subject of an accurate, well-sourced article. Crucially, the yellow box notes that since March 2026, Wikipedia explicitly bans AI-generated article text. You cannot just prompt ChatGPT to write your brand's entry; human editors will flag and nuke it immediately.
• The final slide maps the Wikipedia GEO workflow. If a page doesn't exist, you must generate notability through earned PR first, then submit via 'Articles for Creation' while disclosing employer relationships. For existing pages, propose specific, sourced changes on the Talk page rather than directly editing the article. And importantly, Guis advises treating Wikidata as the more approachable adjacent lever. It's structured data that feeds directly into AI knowledge graphs without the intense editorial friction of Wikipedia.
• That brings us to the end of this Stanford lecture. Three vital takeaways for your own brand. First, stop writing fluffy narrative blogs; transition to answer-first, definitional sentences that an LLM can easily lift. Second, you must author your own comparative positioning, or the AI will pull from your competitors. Finally, treat Wikipedia and Wikidata as your most critical infrastructure, focusing on verifiable facts and original research. Thank you to our experts for breaking this down.