Practical writing on AI search, AI marketing systems, and the parts of SEO that still matter. No hype, one idea per piece, always something you can use.
Start here
What AI search changed, what it did not, and a four-layer framework for finding where your visibility is breaking, with the method to measure it.
AI SEO for healthcare is being the practice or provider assistants name when someone researches a symptom, condition, or treatment. Health is a high-stakes category, so assistants weigh credentials, accuracy, and independent corroboration heavily. The moves: unambiguous service and location signals, genuine clinical expertise on the page, and citable, medically accurate answers.
AI-generated content does not hurt your SEO because a model wrote it. Search systems judge the content, not the tool. It hurts you when it does what unedited AI does by default: restate the consensus with no original data, experience, or judgement. That kind of content is hard to rank and almost impossible to get cited in AI answers, because it adds nothing a model could not already produce.
You rarely win back the exact clicks AI Overviews absorb, because the answer now happens on the results page. What you can recover is visibility: being the source the Overview cites, ranking for queries it does not trigger, and capturing the higher-intent clicks that remain. Confirm what you lost in Search Console, then shift from chasing volume to earning citations and intent.
AI SEO for local business is being named when someone asks an assistant for a service near them. Assistants lean on the signals that feed the local pack plus independent corroboration: a complete Google Business Profile, matching contact details everywhere, and real reviews. The moves: lock entity consistency across directories, and publish extractable, location-specific answers.
AI SEO for professional services is being the firm an assistant recommends when a buyer describes a problem rather than a service. With no product to compare, assistants rely on how clearly you define your expertise and how independent sources describe you. The moves: a sharp specialism entity, demonstrable expertise on the page, and citable, problem-framed content.
An AI visibility score is a single number summarising how often, and how prominently, AI assistants mention or cite your brand across a set of test prompts. It is useful as a trend line, but on its own it is a vanity metric: it tells you that visibility changed, not why or what to fix. Treat it as the thermometer, not the diagnosis.
AI SEO for B2B is staying in the consideration set when buying committees research solutions inside AI assistants. B2B is exposed because research-heavy journeys are full of category, comparison, and best-tool questions that assistants answer directly. The moves: own your category entity, earn analyst and review corroboration, and make solution and comparison pages extractable.
AI SEO for ecommerce is staying visible when shoppers ask an assistant what to buy rather than scan a list of results. Stores are exposed because product, category, and best-X questions are what assistants answer by naming a few options. The moves: clean product and collection entities, structured data on every listing, third-party review corroboration, and extractable comparison pages.
AI SEO for law firms is being the firm assistants name when someone asks for help with a legal problem. Law is a trust-sensitive category, so assistants weigh credentials, a consistent entity, and independent corroboration heavily. The moves: unambiguous practice-area and location signals, real attorney credentials on the page, and citable, jurisdiction-specific answers.
AI SEO for hotels is staying visible when travellers ask an assistant where to stay instead of scrolling a results page. Hotels are exposed because much discovery is now best-hotel-in-X questions that assistants answer by naming a few properties. The moves protect direct bookings: a consistent identity, strong structured data, authentic reviews, and content answering travellers' real questions.
AI SEO for SaaS is staying visible when buyers research software inside AI assistants instead of a list of links. SaaS is unusually exposed, because much of the buying journey is comparison and best-tool-for-X questions that assistants answer by naming a few products. The moves: own your category entity, earn third-party corroboration, and make your product and comparison pages extractable.
Schema markup helps AI systems understand what a page is and how its parts relate, which supports retrieval and entity clarity. It is not a ranking or citation lever, and adding it will not by itself get you into AI answers. The types worth your time clarify identity and content: Organization, Person, Product, Article, FAQPage, and BreadcrumbList.
ChatGPT and Perplexity both answer with cited sources, but they retrieve differently. Perplexity is search-first: it runs a web search on almost every query and cites heavily. ChatGPT blends its trained knowledge with web search when it judges a query needs current information, and cites more selectively. For visibility, the same short list of things gets you cited in both.
Google Gemini grounds its answers in Google Search, so showing up in Gemini is largely the same work as showing up in Google: be retrievable, rank for the query and its sub-questions, and write extractable answers. Because Gemini, AI Mode, and AI Overviews share the same model and index, most of what earns you one earns you the others. There is no separate Gemini placement to buy.
Microsoft Copilot grounds its answers in Bing's search index, so showing up in Copilot depends first on being indexed cleanly by Bing, then on being a citable source: retrievable, extractable, and clearly identified. The most overlooked step is Bing itself, which most marketing teams stopped checking years ago. Copilot cannot cite a page Bing has not indexed, no matter how well it ranks on Google.
AI Overviews are right often enough to be useful and wrong often enough that you cannot trust them blindly. They synthesise from web pages, so they inherit the errors and gaps of the sources they draw on, and occasionally state confident nonsense. For a business there is a second problem that matters more: an Overview can describe your own company inaccurately, and that one you can influence.
Does llms.txt work? Not as a ranking or citation lever, on current evidence. Google says it does not affect Search or AI Overviews, and studies find AI crawlers rarely fetch the file. Where it does work is documentation and agents, helping a tool find the right page to finish a task. It is cheap to add as a low-risk bet, but do not expect it to move your AI visibility today.
An llms.txt file is Markdown at your domain root: an H1 with your site name, an optional blockquote summary, then H2 sections of curated links, each page with a one-line description. A worked example is below. The rule that matters is curation: list the handful of pages you most want an AI to understand, not every URL, and generate the file so it never goes stale.
An AI Overview is generated, not looked up. When a query triggers one, Google breaks it into sub-questions, pulls passages from pages already ranking for those components, and uses a Gemini model to write a single answer with a few cited sources. It is a layer built on top of ordinary search ranking, not a separate index you submit to.
To track AI Overviews, watch three things on a fixed set of buyer queries: whether an Overview appears at all, whether it names or links you, and what it does to your clicks. Search Console will not break this out, so it is part manual observation, part rank tracker, and part a GA4 read of click loss on pages whose position held. The discipline is the same query set, re-run on a schedule.
You track AI search traffic by isolating the referrals assistants send you. In GA4, build a segment or channel matching the assistant domains, chatgpt.com, perplexity.ai, gemini.google.com and the rest, then watch which pages they land on. The catch: this captures clicks from assistants, not citations inside AI Overviews, which send no referral, so it is only half the picture.
An AI SEO strategy is a plan for staying findable as search splits into two outcomes: ranking in a list of links and being cited inside AI answers. It sets objectives for both, allocates effort to whichever layer is actually breaking, and measures rankings and citations separately. It is an extension of SEO strategy, not a replacement, and not the same as using AI tools to do SEO faster.
The generative engine optimization best practices that hold up are principles, not tricks: measure before you change anything, write answers that stand on their own, keep one consistent identity across the web, and earn corroboration you cannot fake. Most of the value is in avoiding the expensive mistakes, chiefly publishing volume and gaming the model, rather than in any single tactic.
Generative engine optimization services help your brand get retrieved, understood, and cited by AI assistants. They cover two jobs: diagnosis, which measures your AI visibility and finds what to fix first, and delivery, the technical and content work that closes the gaps. Start with diagnosis, keep it independent of delivery, and treat any guarantee of placement as a reason to walk away.
Generative engine optimization works by influencing each stage of how an AI engine builds an answer: it retrieves live sources, breaks them into passages, selects the most relevant and trustworthy ones, and synthesises a cited response. You cannot control the output, but you can improve your odds at every stage, from being retrievable to being the clearest passage to quote.
Answer engine optimization (AEO) is the practice of making your content the source an answer engine cites when it responds to a question directly, instead of returning a list of links. Answer engines include ChatGPT, Perplexity, and Google's AI Overviews. AEO is the same core work as SEO, aimed at being named in the answer rather than ranked beneath it.
Generative engine optimization (GEO) is the practice of making your content the source an AI system draws on and cites when it generates an answer, in tools like ChatGPT, Google AI Overviews, and Perplexity. It is the same core work as SEO aimed at a different outcome: being named in the answer, not ranked in a list of links a buyer no longer scrolls.
AI tools for SEO fall into a few categories: content optimisation, keyword and intent research, technical crawling, and the newer AI visibility trackers that show whether AI answers cite you. Each speeds up part of the work, but none replaces the judgement that decides what to fix first. Pick tools by the job you need done, not by the length of a features list.
The generative engine optimization strategies that work match how AI builds an answer: make your content retrievable by AI crawlers, make your identity consistent so a model knows who you are, structure key answers so they extract cleanly, and earn independent corroboration. The winning move is to fix your lowest broken layer first, not to spread effort thinly across tactics.
ChatGPT is useful for the research and drafting parts of SEO: clustering keywords, building content briefs, and drafting titles and meta descriptions. It is unreliable on live facts, and publishing its raw output at scale tends to lower rankings, not raise them. Keep an editor on every output, and never let it invent numbers or sources.
Claude suits the analytical and editorial side of SEO: reading long documents, auditing site structure and content in one pass, clustering search intent, and editing drafts to a consistent voice. Its long context makes it strong for working across a whole content set at once. It should assist judgement, not replace it, and never be trusted with unverified facts.
AI SEO is optimising your content to appear in AI-generated answers, not just ranked links. It means being retrievable by AI crawlers, being a clear entity a model can identify, and being cited when assistants like ChatGPT answer a buyer's question. It shares SEO's foundation but is measured on citation, not position. It is the same practice sold as GEO, AEO, and LLM SEO.
AI visibility is how often, and how accurately, your brand appears in the answers AI assistants like ChatGPT and Google AI Overviews give to buyer questions. You measure it by running a fixed set of prompts and recording whether you are cited, mentioned, or absent. It is the AI-era equivalent of a ranking, and most companies have never checked theirs.
ChatGPT is most useful in marketing as an assistant inside a defined process: research, first drafts, repurposing, and structured analysis, all reviewed by someone who knows the brand and the facts. It is weakest as an autonomous content engine, where it produces confident, generic output. The value comes from the system and judgement around it, not the prompts.
Cost per lead (CPL) is the amount you spend to generate one lead, calculated by dividing campaign spend by the number of leads it produced. A good CPL depends on your industry, deal value, and close rate, not on an absolute figure. The trap is optimising CPL in isolation: a cheap lead that rarely converts costs more per customer than an expensive one that does.
Customer acquisition cost (CAC) is the total sales and marketing spend it takes to win one new customer, calculated by dividing that spend by the number of new customers it produced in the same period. Measured honestly it includes salaries, tools, and overhead, not just ad spend. On its own the figure means little. It only becomes useful next to what a customer is actually worth.
Entity SEO is the practice of optimising content around recognised things, people, organisations, products, and concepts, rather than isolated keyword strings. Search engines and AI systems reason in entities connected through a knowledge graph, not in matching text. The goal is to make your brand an entity engines understand and trust, which decides your AI visibility.
Customer lifetime value (CLV or LTV) is the total profit a customer generates over their entire relationship with you. The basic formula is average purchase value multiplied by purchase frequency multiplied by customer lifespan, then adjusted for margin. It matters because it sets the ceiling on what you can afford to spend acquiring a customer.
The LTV to CAC ratio compares customer lifetime value against customer acquisition cost, showing how many times over a customer repays what you spent to win them. Roughly three to one is the widely used marker of a healthy business. Below that, acquisition is eroding your margins. Well above it, you are usually underinvesting and leaving growth unclaimed.
The marketing funnel stages describe how a stranger becomes a customer, usually grouped as top of funnel (awareness), middle of funnel (consideration), and bottom of funnel (decision). Each stage needs different content. The point of the model is not the diagram: it is finding the specific stage where prospects drop out, so you fix the leak that is actually costing you deals.
An MQL (marketing qualified lead) is a lead whose behaviour suggests interest but not readiness to buy. An SQL (sales qualified lead) has shown enough intent for sales to work it directly. The difference is buying intent, and the point where an MQL becomes an SQL is the handoff between marketing and sales. Defining that point is what stops the two teams blaming each other.
ROAS (return on ad spend) measures revenue generated per unit of ad spend. ROI (return on investment) measures profit against the full cost of marketing, not just the media. The difference is scope: ROAS judges a campaign in isolation, ROI judges whether marketing actually made money. A campaign can post a strong ROAS while losing money once salaries, tools, and margin are counted.
AI in email marketing works best on the high-volume, measurable parts: segmentation, send-time, testing, and first-draft copy. These are pattern problems where AI beats manual effort. It overreaches on brand voice, strategy, and judgement. The rule is the same as everywhere in marketing: give AI the toil, keep the judgement, and check anything that reaches the customer.
CRM data quality is how accurate, complete, consistent, and current your CRM records are. It matters because every downstream system, attribution, lead scoring, automation, reporting, and now AI, is only as good as the data feeding it. Dirty data does not announce itself; it quietly produces wrong numbers and misfired campaigns. Keeping it clean is an ongoing discipline, not a one-off project.
A drip campaign is a fixed sequence of messages on a time-based schedule. Marketing automation is the broader category: any marketing triggered by software, including behaviour-based, branching, multi-channel flows. So a drip is one simple type of automation. The difference that matters is time-based versus behaviour-based: a drip runs the same steps on a timer, automation adapts to each person.
Marketing ROI is revenue from marketing minus marketing cost, divided by marketing cost, as a percentage. The formula is trivial. The hard part is the two inputs: counting the full cost, not just ad spend, and attributing revenue honestly across a long, multi-touch journey. Get those wrong and a precise-looking number misleads you. Treat marketing ROI as a directional guide, not a verdict.
Lead nurturing is building a relationship with buyers who are not ready to buy, guiding them to a decision with relevant content over time. A real strategy is a system, not a fixed email drip: it segments by buyer and stage, tracks readiness through lead scoring, and hands to sales at the right moment. Most nurtures fail by sending everyone the same sequence regardless of fit or stage.
A useful marketing dashboard tracks a short set of KPIs across three stages: acquisition, engagement, and revenue contribution. The common mistake is crowding it with activity metrics like impressions and open rates that show effort, not outcome. Lead with the revenue metrics, pipeline sourced, ROI, acquisition cost, that answer whether marketing is actually working.
Marketing operations is the system between marketing strategy and execution: the people, processes, data, and technology that turn a plan into repeatable, measurable output. It owns the workflow, the tech stack, and the measurement, so campaigns ship on time and the numbers can be trusted. Done well, it is what separates marketing that scales from marketing that just stays busy.
AI agents for marketing plan and carry out multi-step tasks on their own, beyond a chatbot or rule-based automation. The real use cases are high-volume, repeatable work: segmentation, campaign execution, personalisation, and lead qualification. They earn their place there and overreach on judgement and unchecked customer contact. Start from the task, and add oversight before autonomy.
AI search optimization makes your content easy for AI to find, extract, and cite, and builds your presence across the sources they read. It rests on four layers: access for AI crawlers, extractable answers, a clear entity, and originality worth citing. Most citations come from off your own site, so off-site corroboration matters too. It is SEO's foundation aimed at a new target.
B2B marketing attribution assigns credit for revenue across a long, multi-person buying journey. It is hard because B2B breaks the assumptions simple models rely on: cycles run months, a committee decides, and much of the journey is untracked. No model is correct, so fit one to how your buyers decide, set the window to your real sales cycle, and fix the account-level data.
To rank on Perplexity you have to be a source its answer is built from, not a page in a ranked list. Perplexity retrieves relevant pages, summarises across them, and cites the ones it leans on, so it rewards content that is clearly structured, gets to the answer fast, and comes from a recognised source. Be retrievable, extractable, current, and a clear entity, and you become citable.
Google AI Mode answers a query with a conversational fan-out: it breaks your question into many sub-questions, retrieves pages for each, and synthesises one answer. You do not rank in it directly. You show up by being a strong, retrievable source for the query and its sub-questions, with extractable answers and the entity authority a synthesised answer leans on. It is AI Overviews taken further.
LLM SEO is optimising your content so large language models find, understand, and cite it in their answers, rather than only ranking it in a list. It leans less on exact-match keywords and more on clarity, structure, entity precision, and corroboration. In practice it is the same work as GEO and AEO under a new name: strong foundations, extractable answers, a clear entity, and authority.
llms.txt is a plain-text Markdown file at your domain root that gives AI models a curated map of your key pages. It is cheap and low risk to add, but be honest about the payoff: Google has said it does not affect Search or AI Overviews, and there is little evidence yet that AI crawlers read it. Its clearest use today is documentation. Do not expect it to move rankings.
You need an AI marketing consultant when the problem is deciding where AI belongs and how to rebuild the system around it, not when you need hands to run tools. A consultant diagnoses the constraint and designs the approach; an agency or your team executes it. If output is up and results are flat, the gap is usually strategy, not tools.
AI can automate the mechanical marketing work that used to need a person: drafting and personalising at scale, classifying and routing inbound, summarising research, triaging data. The gain is real, but automating without judgement scales mistakes as fast as output. Automate the repeatable steps and keep a person on anything that needs a decision or reaches a customer unchecked.
An AI marketing strategy is a plan for where AI improves specific marketing outcomes, not a list of tools to adopt. Start from the outcome you own, decide which tasks AI genuinely does better, and put a human review layer between the model and anything a customer sees. The teams that see a return treat AI as leverage on a system that already works, not a substitute for the strategy underneath it.
To use AI in marketing without losing your voice, keep the model on drafting and structure and keep a person on the final voice edit, because a model pulls copy toward the generic average it was trained on. Write down what your voice actually is, feed it real examples, and treat every draft as raw material a human finishes. The voice is a judgement call you cannot delegate to the tool.
SEO is not dead, but the claim is not pure hype. Being found now has two layers: ranking in the classic results, and being cited in the AI answers above them. The fundamentals that made you rankable still matter, because AI answers are built on top of them. What dies is treating a ranking as the finish line; the work now is being the source the answer is built from.
AI lead scoring ranks leads by likelihood to convert, learning the pattern from your historical data instead of points you assign by hand. That is more accurate when the data is good and misleading when it is not. It is worth adopting once you have clean, outcome-labelled history and sales and marketing agree on what a good lead is. Without both, it just automates a definition nobody trusts.
Multi-touch attribution splits credit for a conversion across the touchpoints that led to it, instead of giving it all to the last click. No model is objectively correct: attribution is a simplifying assumption, not a measurement, so the right model is the one that fits your decision. Treat the numbers as directional, and fix the first-party data underneath before trusting any of them.
Sales and marketing alignment means both teams share one definition of a good lead and one view of the funnel, so demand does not leak at the handoff. It is a systems problem, not a relationship one: the friction comes from misaligned definitions, data, and incentives, not personalities. Fix those and the metrics each team is measured on, and co-operation follows.
In the age of AI, SEO's fundamentals did not stop mattering; they became the precondition for being retrievable. Crawlability, structure, internal linking, and authority are what let an AI find, read, and trust your pages before it can cite them. What changed is the goal: not only ranking a page but being the source an answer is built from, which rewards extractable answers.
Technical SEO for AI crawlers is mostly the technical SEO you know, aimed at new visitors: the bots that build AI search indexes and fetch pages for assistants. They need what classic crawlers need, reachable URLs, fast responses, and content in the server-rendered HTML rather than painted in by JavaScript. Get your content into the initial HTML and let these crawlers in, or you are invisible.
Revenue operations unifies the systems and data behind marketing, sales, and success, so the revenue engine runs as one rather than three teams optimising separately. It closes the handoffs between them, where revenue quietly leaks. You need it when the handoffs break, when teams cannot agree on the numbers, or nobody owns the funnel end to end. It is an operating model, not a tool.
In content marketing, AI belongs on the mechanical work, research, first drafts, restructuring, and variation, not on the parts that make content worth reading: the angle, the judgement, the first-hand experience. Use it to clear the friction that stops good ideas getting written, and keep a person on what is worth saying. Used the other way round, it just makes more forgettable content.
Apply AI to the steps inside a workflow that are repetitive and mechanical, and route it through a person where a decision or customer-facing output happens. The unit is the step, not the whole process: most workflows mix both, so ask which steps to automate, not whether. Map the workflow, mark each step as toil or judgement, and automate the toil.
AI marketing produces more drafts, not more revenue, because output was rarely what limited revenue. Pipeline is constrained by positioning, distinctiveness, and the system that routes and follows up on demand, none of which a model improves by generating faster. Cheap output on an unchanged constraint adds cost, not results. Revenue moves only when you fix the real bottleneck first.
Answer engine optimization services help your brand get found, understood, and cited by AI assistants. They split into two offers: diagnosis, which measures your AI visibility and tells you what to fix and in what order, and delivery, which does the technical and content work. Most should start with diagnosis: paying for delivery before you know the cause spends budget on the wrong layer.
If AI describes your company vaguely, out of date, or as the wrong business, the problem is entity understanding: it has no clear picture of who you are. It comes from how consistently you are described across your site, structured data, and third-party sources. When those disagree, you become a fuzzy entity and a model prefers a clearer competitor. The fix is consistency, not more content.
A generative engine optimization checklist works best ordered by cause: the fixes are prerequisites for each other. Work in this order: make your content retrievable, make your identity consistent so a model knows who you are, make your key answers extractable, earn corroboration, and measure on a fixed schedule. Doing these out of order wastes effort on layers that cannot yet pay off.
GEO and SEO share most of their technical foundation, so you rarely need a separate agency or team for AI search. What you need is a provider who measures the right outcome and has reprioritised for it. Bring in a specialist only when your current provider cannot show a method for measuring AI visibility, or treats AI search as a product to upsell rather than a shift in the work they already do.
To check whether your brand appears in AI answers, run a fixed set of buyer questions through each assistant they use, in a clean session with personalisation off, and record whether you are cited, mentioned, or absent, and how you are described. Check each surface separately: ChatGPT, Gemini, Perplexity, and Google AI Overviews retrieve and answer differently, and one does not predict another.
You do not rank in an AI Overview directly. Google builds it from pages already ranking for the query, so the route in is to earn strong conventional visibility for the query and its sub-questions, then make your answer easy to extract: a direct response near the top, self-contained passages, and coverage of the sub-questions the Overview draws on. There is no separate Overview ranking to buy.
To show up in ChatGPT search, your pages must be reachable by its crawler, present in the index it queries, and clear enough to extract. Four things do it: let OpenAI's crawlers fetch your site, state your answers plainly near the top of each page, keep your name and category consistent everywhere, and earn mentions from sources ChatGPT trusts. There is no submission form and no paid placement.
An AI visibility audit measures how AI assistants find, describe, and cite your brand, and diagnoses why. A thorough one checks four things: whether your pages can be retrieved and rendered, whether your identity is consistent enough to understand, whether your content is extractable, and how you are described and cited across the assistants buyers use. It is a diagnosis, not a score.
AI answers cite the source that adds something others do not. When every page in your category repeats the same advice, a model has no reason to prefer yours, so it draws on whoever supplied the original material: first-hand data, a named method, the primary source. To be citable, stop restating the consensus, publish something only you could have written, and structure it to be extracted.
An assistant recommends the company it can most confidently connect to a category. A competitor gets named instead of you when the model has stronger evidence for them: they are retrieved more often, their pages answer the question directly, their identity is consistent, and other sources corroborate them. It is an evidence gap, not a quality gap, and each cause is diagnosable.
Publishing more content usually lowers AI visibility, because volume works against what these systems reward. A flood of thin, similar pages dilutes your entity, so a model is less sure what you are known for; it repeats the consensus, so none of it is worth citing; and it buries your genuinely strong pages. The fix is fewer, more original pages, and pruning the rest, not publishing faster.
Schema markup helps AI parse a page and connect it to a known entity, but it does not force a citation, override a confused identity, or lift a page with no clear answer in it. It is a comprehension aid, not a ranking lever. If schema is not moving your AI visibility, the problem is usually upstream: retrieval, entity consistency, or extractable content the markup describes but cannot replace.
SEO is optimising to appear in search engine results. AEO, or answer engine optimisation, is optimising to be the source an AI-generated answer draws on. GEO, or generative engine optimisation, means substantially the same thing as AEO and the two are used interchangeably. In practice these are not three separate disciplines but one discipline with a broader definition of where visibility happens.
Check your robots.txt for directives targeting GPTBot, ClaudeBot, PerplexityBot, Google-Extended and similar agents. Blocking them keeps your content out of AI answers that would otherwise cite you. Some agents control training use and others live retrieval: blocking training does not affect whether you are cited today, while blocking retrieval removes you entirely.
AI assistants that browse the web run a search, retrieve candidate pages, extract passages that address the question, and compose an answer from those they judge most relevant and reliable. Being citable therefore depends on three things in sequence: being retrieved at all, containing a passage that directly answers the question, and carrying signals that the source is trustworthy.
Build a fixed set of prompts covering how people describe your category, your brand, and the topics you claim expertise in. Run each in a clean session with memory and personalisation off, several times, across the assistants your buyers use. Score every result as cited, mentioned, or absent, and record which competitors appeared instead. Repeat on a fixed schedule.
If your positions are steady but your clicks are falling, the likely cause is that AI-generated answers are satisfying searches that used to send you a visit. The query still happens, your page still ranks, but the click never occurs. This is measurable in Search Console: find pages where impressions hold or rise while clicks fall, showing as a falling click-through rate at a stable position.
A migration protects search rankings through redirects, but redirects do not transfer citations in AI answers. Assistants cite URLs they have seen, and third-party pages still point at your old ones. Baseline your AI visibility before you migrate, keep every redirect a single hop to a page that answers the old page's question, and re-run the same prompts for 90 days after launch.
Newsletter
What changed in AI search, what it means for marketing teams, and what I would do about it. No summaries of other people’s posts, and no more than one email a week.
Written by Viken Patel. No ghostwriters, no AI-generated filler. Unsubscribe any time.