SEO With AI: Search Intent, Keyword Research, On-Page, the Content Brief, E-E-A-T, and Your First Look at the AI-Search Shift

Let's run the news, fast, because there's a lot moving in AI search this week.

Top item, June 9. Anthropic shipped Claude Fable 5, its most powerful model available to the general public. The pricing that matters to you is ten dollars per million input tokens and fifty dollars per million output tokens, which is reportedly less than half the price of the previous top-tier preview. Here's the part with a deadline. Fable 5 is free on the Pro, Max, Team, and seat-based Enterprise plans from June 9 through June 22. On June 23 it leaves those plans and starts billing against usage credits. So if you want to try it on a real content or research workflow, do it before June 22 while it's free, and mark June 23 as the billing cliff. A small number of sensitive queries, reportedly under five percent, get routed to an older model for safety.

Second item, and this one is the big SEO story. On June 3, Google launched a toggle inside Search Console that lets you opt your site out of AI Overviews and AI Mode while still showing up in normal search. It also shipped a new generative AI performance report that finally shows you impressions inside AI answers and which of your pages surface there. That data simply wasn't available before. The catch: the opt-out is all or nothing. You'd give up your AI-answer impressions to maybe protect your clicks. It also does not cover the Gemini app, so your content can still show up there. The toggle goes live June 17, so do not flip it before then, and open that new report first to see what you'd actually be giving up. This rolled out in the UK first under a binding competition order. For context on why people care: Pew found only eight percent of users clicked a link when an AI Overview was present, versus fifteen percent without one, and publisher referrals are reportedly down about a third.

Now the AI-search snapshot, which is your standing fixture. Similarweb's latest numbers on generative-AI traffic share show ChatGPT down to about fifty-three percent from seventy-six a year ago, Gemini up to twenty-seven percent, more than one in four visits, Claude tripled to about nine percent, and Perplexity near one percent. The takeaway: optimizing for ChatGPT alone leaves real share on the table. Make sure your visibility tracking covers Gemini and Claude too. Rounding it out, AI Overviews now reportedly appear on about a quarter of Google searches, double a year ago. Zero-click runs around forty-three percent when an Overview shows, and about ninety-three percent inside AI Mode. And ranking still feeds AI: position one earns roughly a one-in-three citation chance, dropping to about one in eight at position ten. Treat all of these as directional. That's the rundown.

Okay. This is the episode a lot of you have been waiting for. We're going to do SEO with an AI assistant, end to end, and then take a first look at how AI search is changing the whole game. By the end you'll know how search actually works, how to read search intent, how to do keyword research with AI without getting burned, how to handle the on-page details, how to write the one document that ties it all together, and what to do today about AI answers eating your clicks. Let's build it from the ground up.

Start with how search works, glossed for people who don't write code, because everything else sits on top of this. There are three steps. The first is crawl. Google runs an automated program, you'll hear it called Googlebot, or a crawler, or a spider, and it follows links from page to page and fetches the HTML it finds. The second step is index. Whatever it fetched gets stored and organized in Google's enormous database. This matters more than people think: if a page is not in the index, it cannot appear in search, full stop. You can check whether a page is indexed by typing site colon your domain slash the page into Google. If it shows up, it's indexed. If it doesn't, you've got a problem before you've even started. The third step is rank. Out of all the indexed pages, Google's algorithms sort them by relevance and quality and arrange them into the results page. That results page has a name we'll use all episode: the SERP, which stands for search engine results page.

Sitting on top of crawl, index, and rank is one more idea: intent. Google doesn't just match the words you typed. It tries to infer why you searched, and then serves the format that matches that why. Classic SEO is about earning what people call a blue link, one of the ten organic results on the page. Here's the shift we're going to keep coming back to. AI search changes the unit of victory. It used to be: rank a blue link. Now it's increasingly: get cited inside the answer that Google or a chatbot writes for the user. But notice this. Crawl, index, and rank still run underneath both worlds. So the foundations do not go away. That's the reassurance to hold onto while everything around it changes.

Let's talk about search intent properly, because intent governs everything downstream. There are four types. The first is informational. The person wants to learn something. The language gives it away: how, what, why, guide, tutorial. This is the biggest bucket, roughly seventy percent of all searches. The second is navigational. The person already knows where they want to go and is using search to get there. Think someone typing OCDevel podcast, or Mailchimp pricing, because they want that specific site. The third is commercial, sometimes called commercial investigation. The person is researching before they buy. The tells are words like best, top, versus, review, alternative. Something like Semrush versus Ahrefs is pure commercial intent. The fourth is transactional. The person is ready to act right now. The language is buy, pricing, coupon, near me, free trial, demo.

Here's why intent is not academic. Intent governs the page format Google will rank. If someone searches a transactional buy query and you've written a long how-to guide, you have mismatched the intent and you will not rank, no matter how good your guide is, because Google wants to show a product or pricing page in that slot. So before you write a single word, you ask: what does the searcher actually want here, and what format satisfies that want?

Now connect intent to your funnel, because this is where marketers consistently waste money. Top of funnel, the awareness stage, maps to informational intent. The assets are blog posts, how-to guides, explainers. Middle of funnel, the consideration stage, maps to commercial intent. The assets are comparison guides, the X versus Y pages, case studies, reviews. Bottom of funnel, the decision stage, maps to transactional intent plus your brand navigational searches. The assets are pricing pages, product pages, free-trial and demo landing pages, testimonials. Quick vocabulary so we can speak in shorthand: top of funnel is TOFU, middle is MOFU, bottom is BOFU.

Here's the mistake almost everyone makes. They over-index on top-of-funnel blog posts, because those are cheap and easy to write, and they starve the bottom-of-funnel pages that actually convert. Twenty informational blog posts feel like progress, but if your money keyword is commercial or transactional, you've built traffic that never buys. This is exactly why intent classification matters. It stops you from writing twenty awareness posts when the page that pays the bills is a comparison or a pricing page.

Now let's do keyword research with AI, and I'm going to give you a real workflow plus the one hard limit you must never forget. Step one is seed terms. Pull in the brand-voice profile and the audience research you built in earlier episodes so the AI knows who it's writing for. Then you'd tell it something like this: I sell this product to this audience persona. Act as an SEO strategist. Generate fifty seed keyword ideas this persona might search across their buying journey. Group them into awareness, consideration, and decision. For each one, label the intent, informational, navigational, commercial, or transactional, and tell me whether it's a head term or long-tail. That single prompt gives you a structured starting map instead of a blank page.

Step two is long-tail expansion. Quick definition: a long-tail keyword is a longer, more specific phrase. It has lower search volume, but it's far more relevant, it converts better, and it has way less competition than the short head terms everyone fights over. So you'd take one head term and tell the assistant: take email marketing, and expand it into thirty long-tail variations a small-business owner would actually type, including question forms, comparisons, and best-X-for-Y phrasings. Now you've got specificity you can rank for.

Step three is intent classification and clustering, and this is where AI is genuinely excellent. It reads the linguistic and semantic patterns and groups keywords by shared intent even when the wording is totally different. You'd hand it your list and say: here are two hundred keywords, cluster them into groups where each group could be satisfied by one single page, name each cluster, list its keywords, identify the dominant intent, and suggest one primary keyword per cluster. Let me gloss the professional version of this so you know what the AI is approximating. It's called SERP overlap, or SERP-based clustering. The idea: two keywords belong on the same page if the same URLs rank in the top ten for both of them. If they share seventy percent or more of those ranking URLs, that's a strong signal they're really one topic. The AI guesses at this from language. Only a real tool can see the actual SERPs and confirm it.

And now the hard limit, which is half of today's big closing lesson, so burn it in. A general AI assistant does not know real search volume. It does not know keyword difficulty. It does not know cost per click. If you ask it the monthly search volume for some keyword, it will confidently hand you a specific number, and that number is hallucinated. It's the same slop tell we keep hitting: fluent, specific, and wrong. You must validate every volume in a real keyword tool. Let me name three. Google Keyword Planner is free, it lives inside Google Ads, and it gives you volume in ranges like one thousand to ten thousand. Semrush is a paid suite, around a hundred and forty dollars a month to start, and it gives precise volume, a keyword difficulty percentage, and SERP features. Ahrefs is also paid, with a huge database, its own keyword difficulty score, and a parent topic feature. The division of labor is the whole point. The AI generates and organizes the ideas. The keyword tool supplies the real numbers. You make the decision.

Let's move to on-page work with AI, and here the exact limits are load-bearing, so I'm going to be specific. First, the title tag. That's the clickable headline in the search results and the text in your browser tab. Target fifty to sixty characters, but the real limit is pixel width, about six hundred pixels on desktop, so stay around five hundred eighty to be safe, and mobile truncates around fifty-five characters. Put your primary keyword near the front. Your one H1 heading on the page is essentially your on-page title. You'd prompt: write five title tag options for a page targeting this keyword, each under sixty characters, keyword near the front, human, no clickbait, and show me the character count after each one.

Next, the meta description. That's the snippet of text under the title in the results. It is not a direct ranking factor, but it drives click-through, and Google often rewrites it anyway. Target a hundred fifty to a hundred sixty characters, around nine hundred twenty pixels on desktop, but on mobile you've got roughly a hundred twenty characters, so front-load your message. You'd say: write three meta descriptions, a hundred fifty to a hundred sixty characters each, include this keyword once, name the benefit, add a soft call to action, and show the counts.

Headings come next. You want exactly one H1. Your H2s mark the major sections. Your H3s nest underneath them. Good heading structure helps human readers, and it helps AI engines parse and extract your content, which matters more every month. Then internal linking, which is links from your own pages to your other pages. These help Google discover and crawl your site, they pass relevance and authority around, and they help readers go deeper. AI is great at suggesting anchor text and link targets, and the practical habit is to link any new piece of content from your existing pages quickly so it gets found.

Image alt text is next. That's a text description that screen readers speak aloud for blind users and that search engines read. Keep it concise, accurate, and in-context, and do not stuff keywords into it. Google explicitly warns against that. AI vision can auto-draft alt text by looking at the image, and then you edit it. You'd prompt: write alt text for this image, under about a hundred twenty-five characters, describe what's shown for a blind user, and include the keyword only if it's genuinely relevant.

Last on-page piece, schema, also called structured data, and I'll keep this a gloss. It's a bit of code, and the format that matters is called JSON-LD, that tells search engines what your content actually is: an article, a product, a recipe, a set of FAQs. It can earn you rich results, and it helps AI engines extract your content cleanly. The types worth knowing in 2026 are Organization, Article or BlogPosting, Product, LocalBusiness, and FAQPage. One important caveat: Google deprecated FAQ rich results in regular search back in May 2026. But here's the twist that makes it still worth doing. AI engines still crawl, extract, and cite FAQ structured data. So it shifted from being a rich-snippet play to being an AI-citation play. Same code, new reason.

Now the keystone of the whole episode: the content brief. This is the document that ties everything together, the keyword, the intent, the outline, and your quality signals, into one artifact you can hand off. Hand it to a freelancer, hand it to future-you who's forgotten the plan, or hand it to an AI generation run. And here's the 2026 update. You write the brief as structured fields, not loose prose, because the exact same brief now governs both a human writer and an AI draft, and machines need fields.

Let me walk the fields, because this is the recipe. It's one to three pages. Your primary keyword, exact, plus your secondary keywords. The intent classification and the matching SERP format. The target persona. The required headings, your H2s and H3s, drawn from the intent and from gaps you spot in competitors. The entities, meaning the specific concepts and brands the page must mention to read as topically complete to a search engine. A word-count range, and this is key, set by what's actually ranking for this query, not by some house standard you invented. Your evidence and sources: the data, stats, quotes, and first-hand examples, because that's the fuel for quality. Your quality signals, like a named author with credentials and genuine first-hand experience. Your on-page specs: title, meta, URL, schema type, internal links. And your conversion plan: the primary and secondary calls to action.

You can have AI build the skeleton. You'd say: act as an SEO content strategist, build a content brief for this keyword, the intent is this type and the persona is this. Output a primary keyword plus five secondary keywords, the intent, a suggested H1 and an H2 and H3 outline, eight entities the page must mention, a target word-count range based on what ranks for this query, three questions the page must answer directly, and two quality signals to add. And then the crucial instruction: flag anything you're unsure about rather than guessing. Here's why the brief is so powerful. The content brief is your slop firewall. If the brief demands first-hand evidence, a named expert, and specific named sources, then the AI draft physically cannot come out generic, because you've required the things that make it specific.

That brings us to quality, and the framework Google uses, which is E-E-A-T. Four letters, four pillars: Experience, Expertise, Authoritativeness, and Trustworthiness. It lives in Google's Search Quality Rater Guidelines, the document human raters use to score results. Those raters don't change rankings directly. Their scores train the systems that do. One bit of history worth knowing: the first E, Experience, was added in December 2022. Before that it was just E-A-T, expertise, authoritativeness, trust.

Let me define each pillar, because the gap between two of them is the whole point. Experience means first-hand involvement: you actually used the product, you actually visited the place. Expertise means knowledge and credentials. Authoritativeness means you're a recognized go-to source on the topic. And Trustworthiness, which Google calls the most important of the four, means accuracy, transparency, and safety. Now the load-bearing insight. An AI has read about everything and experienced nothing. It can simulate expertise beautifully, because it can summarize what every expert has said. But it cannot supply Experience. It cannot give you your actual test results, your screenshots, your customer's story, your I-ran-this-for-ninety-days-and-here's-what-broke. That first-hand layer is the human's irreplaceable contribution, and honestly it's the single thing that separates an AI draft from publishable content. The trust signals that show it: author bylines with real credentials, a proper about page, citations to primary sources, visible dates, and contact info.

Now the pitfall everyone worries about: AI content and Google's actual stance, and let me give you the dated facts so you're not operating on rumor. Google has not banned AI content. I'll say that clearly. Google evaluates quality regardless of whether a human or a machine wrote it. AI content is completely fine if it adds real value. It gets the lowest possible rating if it adds zero value. The dated event that defines this is the March 2024 core update and its new spam policies. Two big things happened. First, the helpful content system got folded into the core ranking algorithm, so it's now a continuous, real-time signal instead of a periodic standalone update you could wait out. Second, three new spam policies landed.

Policy one is scaled content abuse, and this is the one marketers actually trip. Google defines it as many pages generated mainly to manipulate rankings rather than help users, and they explicitly include using generative AI to produce many pages without adding value. That's the mass-published AI article problem in one sentence. Policy two is site reputation abuse, sometimes called parasite SEO, which is putting third-party content on a trusted host site to exploit its ranking authority. Policy three is expired domain abuse. Now hear the unifying principle, because it's what keeps you safe. Every one of these policies hinges on primary purpose to manipulate rankings rather than help users. Automation is not the trigger. AI is not the trigger. Intent to manipulate at scale with low value is the trigger. On enforcement: the March 2024 update, then a February 2025 update, then an August 2025 spam update all kept cutting low-quality content, and Google's anti-spam engine, which is itself AI-based, is called SpamBrain. Google even gives you a self-audit you can copy. Three questions. Is it self-evident who created this? How was it made, and if AI was involved, is that disclosed? And why does it exist, to help people or just to rank? If a draft is fluent but sourceless, has no first-hand detail, hedges in generic phrasing, or has fabricated stats, and if it could be about literally any brand, that's thin content waiting to get down-ranked.

Okay. Now the part you came for: a first look at the AI-search shift. And I want to be honest about scope. This is a taste, not the full system. The complete generative-search playbook is its own Act II episode. The core change is simple to state. Searches increasingly end without a click, because the answer is generated right there on the page or inside a chatbot. So the goal shifts from rank a blue link to get cited inside the answer.

Some numbers, and treat these as fast-moving, verify them yourself. Roughly sixty-eight percent of Google searches end without a click in early 2026, up from about forty-five percent a decade ago. When an AI Overview is present, the zero-click rate averages around eighty-three percent, versus about sixty percent without one. AI Overviews show up on a large and growing share of searches, reports range from twenty to forty-eight percent depending on method and vertical, and when they appear they cut organic click-through by around sixty percent. But here's the nuance that should change your strategy. The variance by vertical is enormous. How-to and informational queries get cannibalized almost completely, around ninety-nine percent. But e-commerce queries, the buy-X and best-X and price-of-X searches, trigger AI Overviews only about three or four percent of the time. So your top-of-funnel informational traffic is the most exposed, and your bottom-of-funnel transactional pages are relatively shielded. That's a strategic gift, because it tells you where to spend your effort.

Let me name the players so the vocabulary doesn't trip you. Google AI Overviews, the AI answer box at the top of the results, often shortened to AIO. Google AI Mode, the conversational tab inside Search that launched in 2025 and has now passed a billion monthly users. ChatGPT search, from OpenAI, which started as a prototype called SearchGPT in July 2024 and went generally available as ChatGPT search at the end of October 2024. Perplexity, an answer engine founded in 2022 that runs a real-time web search on every single query and gives you inline numbered citations. And Gemini, which is Google's assistant and the model powering AI Mode.

Now the alphabet soup, and I'll be straight with you about it. GEO stands for generative engine optimization, optimizing so that large language models cite you as a source. AEO stands for answer engine optimization, structuring your content so answer engines extract and cite it. Honestly, in early 2026 there's no settled consensus on the distinction. The field also throws around GSO and LLMO. My advice: treat GEO and AEO as roughly interchangeable, and read both as getting your brand into AI answers.

Here's how getting cited differs from ranking a blue link, mechanically. AI engines use something called query fan-out. They break your one question into several sub-queries and pull from many different pages to assemble the answer. So you can get cited without ranking number one. Ranking still matters, but less deterministically. Most AI Overview citations, somewhere around eighty-four to ninety-two percent, do include a domain from the organic top ten. But one 2026 analysis found that the share of citations coming from the top ten dropped from about seventy-six percent down to thirty-eight percent, which means a lot of citations now come from pages ranking eleven to a hundred or beyond. So what actually correlates with getting cited? Semantic completeness, meaning you covered the topic thoroughly. Clear structure: headers, concise quotable paragraphs, and a direct answer right up top. Entity richness. And third-party corroboration, meaning your brand gets mentioned across independent places like Reddit, YouTube, G2, and industry press. AI looks for agreement across independent sources, and YouTube mentions in particular correlate strongly.

One more reality check: the platforms cite very differently from each other. A 2026 study of over thirty-four thousand responses found ChatGPT named brands in only about half a percent of responses, while Perplexity did it around thirteen percent of the time, and Perplexity averaged around twenty-two citations per response. ChatGPT leans heavily on Bing's top ten. And only about eleven percent of the domains cited showed up in both. So they are genuinely different surfaces.

Here's the reassurance, and it's the through-line of the entire episode. Classic SEO still matters, because crawlability, indexing, structure, schema, topical authority, and E-E-A-T are the exact same signals that get you cited inside AI answers. Answer-engine and generative optimization is largely your SEO foundations, plus strong structure, plus off-site reputation. It is not a replacement for what you already learned today. The foundations carry forward.

So what's the one thing to do today? Run your real customer queries through the AI assistants and see if and how your brand shows up. Here's the workflow, and it's manual on purpose, because that's the right scope for a first look. Take ten to twenty actual bottom-of-funnel and middle-of-funnel queries. Things like best category-tool for your-audience, or your-brand versus a-competitor, or is your-brand good for this use case. Type each one into ChatGPT with search turned on, into Perplexity, and into Google AI Mode or AI Overviews. For each, note four things. One, are you mentioned at all? Two, are you cited with an actual link? Three, is what it says about you accurate? And four, which competitors and which third-party sources, the Reddit threads, the G2 pages, the listicles, get cited instead of you? That gap list is your first generative-optimization to-do list.

You can also turn the assistant on itself. Type this into it: when someone asks you to recommend the best tool in my category for my audience, which specific products and brands do you name, and what sources are you drawing on, and where would my brand fit and why? The answer tells you exactly what the model believes and where the holes are. There are paid trackers that automate all of this, you'll hear names like Profound, Otterly, Semrush's AI Visibility Toolkit, and Ahrefs Brand Radar, and they monitor how your brand shows up across all the assistants. But for a first-look episode, the manual run-your-own-queries method is exactly the right scope. Do that by hand first.

Let me close by tying the two big pitfalls together, because they're really two faces of one mistake, and they're the whole lesson. Pitfall one: treating AI keyword volumes as real. A general assistant will hand you confident, fabricated volume and difficulty numbers. Validate every volume in Keyword Planner, Semrush, or Ahrefs before it drives any strategy. The cue is simple. If the number came out of a chat box instead of a keyword tool, it's a guess. Pitfall two: publishing scaled AI content that trips the helpful-content and scaled-content-abuse signals. The cues are generic copy that could be any brand, no first-hand experience, no named author, no original data, all mass-published in a short window. The fix is brief-driven drafts that require real evidence, a named expert or editor, real examples, and human editing, with AI as the drafting tool and never the publishing button.

Here's the through-line to carry into the next Act. AI is the fastest research and drafting partner you've ever had. But it knows no real numbers, it has no real experience, and it publishes no real authority. The marketer supplies all three. And the foundations we built today, intent, structure, the content brief, and E-E-A-T, are exactly what carries you into the AI-search era that the next Act turns into a full system. Go run your queries. That's your homework.