In this blog post I'm going to walk you through how to do AI SEO in 2026. Specifically, how to get cited, summarised, and recommended by ChatGPT, Claude, Perplexity, and Google AI Overviews on the queries that bring you real buyers. Not the version where you stuff your post with FAQ schema and pray. The one I've been running on my own site for the last nine weeks of a public rebuild, with screenshots, real numbers, and a fair amount of swearing.
Most of what you'll read about AI SEO this year was written by someone who hasn't watched their own traffic graph cliff-edge in the last six months. They've got a Notion checklist, three plugin recommendations, and a take.
The take's wrong.
The bit nobody tells you is that AI SEO in 2026 is mostly the unglamorous SEO work you've been avoiding for three years, plus a small amount of new thinking about how language models pick sources. That's it. The new thinking matters. But it's the unglamorous work that moves the needle.
By the end of this blog you'll know exactly what AI SEO is, why it isn't the same thing as regular SEO, how the four main AI engines pick content to cite, what to do this week on your own site to get found, and what most people get spectacularly wrong. You'll also see what nine weeks of doing this on my own site has produced. The good, the bad, and the bit where I de-indexed 1,300 pages on a Sunday.
TL;DR
AI SEO in 2026 is the work of getting your content cited, summarised, and recommended by ChatGPT, Claude, Perplexity, and Google AI Overviews. The fundamentals are still on-page SEO, technical health, topical authority, and original first-hand information. What's new is structuring content so it survives extraction by language models, and removing the pages that drag your domain down.
Why AI SEO matters more than Google SEO right now
Here's a number that should worry you. In the last twelve months, the share of search queries that get answered by an AI overview, an AI assistant summary, or a chat-style answer (rather than ten blue links) has gone from a margin of search behaviour to roughly a third of all consumer queries, depending on whose data you trust. ChatGPT alone now handles hundreds of millions of weekly queries that would have been Google searches three years ago.
That is a permanent change in how people find information. It is not a fad. It is not the next Clubhouse.
If you're a business owner, marketer, or founder relying on Google traffic for leads, your traffic chart at the end of 2026 will look one of two ways. Flat-to-up because you've adjusted, or down forty percent because you haven't.
(That's the worst-case I'm seeing in my client audits. Small sample, but consistent.)
Three things have happened at once.
One, Google itself started putting AI Overviews above the organic results on a meaningful chunk of queries. Click-through rates to position one collapse when an AI Overview answers the query first.
Two, ChatGPT and Claude both ship with browsing now. People ask them questions, the model goes off, fetches a few pages, summarises the answer, and cites a couple of sources. If you're not in the cited sources, you're invisible, even if you'd have ranked first on Google for the same query.
Three, Perplexity has eaten a real chunk of the research-style queries that used to live on Google. Anything that starts "what's the best...", "compare...", or "how does X work..." has a meaningful chance of being asked in Perplexity now, and Perplexity has its own ranking logic.
The blunt version. The Google-only SEO playbook works on a smaller and smaller share of queries every quarter. AI SEO is what works on the rest of them.
Good news? Most of the AI SEO playbook is the SEO playbook you should already be running, with a few extra layers on top. Bad news? The few extra layers are the bits people skip, and they're the bits that decide whether the AI engines cite you or your competitor.
The bit nobody selling you AI SEO services wants you to know. The foundations matter more than the new stuff. If your site is a mess of thin content, broken internal links, and pages that haven't been touched since 2019, no amount of AI-friendly structuring will save you. I had to learn that the hard way, on my own site, this autumn. (More on the rebuild later.)
The takeaway. AI SEO isn't a separate discipline. It's regular SEO done properly, plus a handful of structural changes that help language models extract your answers cleanly. Skip the foundations and the structural changes do nothing.
What is AI SEO and how is it different from regular SEO?
Definition first.
AI SEO is the work of optimising your content, your site structure, and your wider footprint on the internet so that AI engines (ChatGPT, Claude, Perplexity, Google AI Overviews, and the rest) recognise you as a source worth citing, summarising, or recommending when someone asks them a question in your niche.
Regular SEO is ranking in the organic search results on Google. Same goal, get found, but the mechanism is a list of links sorted by Google's algorithm, and the user clicks through.
The difference matters because the success criteria are different.
In regular SEO, the click is the win. You rank, the user clicks, you get a visit. Conversion is downstream of the click.
In AI SEO, the citation is often the win, even if the user never clicks through to your site. ChatGPT or Claude reads your post, summarises the answer, and either credits you in the response or doesn't. Most of the time the user gets the answer they need without leaving the chat. The traffic doesn't move. The brand recall does.
That sounds horrifying if you measure success by sessions in Google Analytics. It's the wrong frame.
Two reasons.
One, the share of citation traffic that does click through converts at a much higher rate than organic Google traffic, in my client data. The user has had the answer summarised, asked a follow-up, asked another follow-up, and is now coming to your site as a half-warmed lead.
That's a different conversation to a cold Google visitor.
Two, brand recall built through being cited dozens of times per week by AI engines compounds in a way Google ranking never did. People start asking ChatGPT, "what does Lilach Bullock say about X?" That doesn't happen by accident. It happens because the model has summarised you so often it's stitched your name into the answer.
So AI SEO isn't a smaller version of regular SEO. It's a different game with overlapping rules.
The foundations both share
Site speed. Mobile usability. Clean internal linking. Strong topical clusters. Original first-hand information that isn't a regurgitation of what's already on the first page of Google. Schema markup. Author bios with E-E-A-T signals. The whole thing.
If your site fails on any of these, AI SEO is going to fail too. The AI engines are pulling from the same web Google indexes. They're applying their own filters on top. If Google won't rank you, the AI engines mostly won't cite you either.
The bits that are AI SEO specific
Three things matter here that don't matter as much in regular SEO.
First, structural extractability. Can a language model read your post and pull out a clean, quotable answer in two to four sentences? If your content is a wall of meandering paragraphs that buries the answer in section seven, no, it can't. The AI engines summarise the first thing that looks like an answer. Train your content to put the answer in front, every time.
Second, entity richness. AI engines build a graph of named things. People, tools, brands, frameworks, dates. Content that names specific entities (the 2024 HubSpot State of Marketing report, or Claude Opus 4.7, or Greenpeace) gets pulled in. Content that says "a major report" or "the latest model" doesn't.
Third, source diversity around your name. The AI engines triangulate. They want to see your name and topic showing up across multiple credible sources, not just your own site. Podcasts, guest contributions, mentions in roundups by other people, your own LinkedIn posts. If the only place anyone's heard of you is your own site, the model is going to be cagey about citing you. (This is the bit most solo founders hate.)
The takeaway. AI SEO shares foundations with regular SEO, but lives or dies on three things on top. Extractable answers. Named entities. Distributed authority.
How do AI engines pick which content to cite?
Plain answer first, then nuance.
AI engines pick content to cite based on three layers stacked on top of each other. Underneath is the model's training data, which gave it a baseline understanding of who's credible in any given topic. On top of that is the live retrieval layer, where the model fetches a handful of pages from the web in real-time. On top of that is a ranking step where the model decides which of those pages to summarise, which to cite, and which to ignore.
You can't influence the training data layer in any short timeframe. (The next training cut-off will eventually catch up to whatever you publish today, but we're talking many months.) The retrieval and ranking layers are where the work happens.
ChatGPT
ChatGPT with browsing on uses Bing's index for retrieval. So if you don't rank on Bing, you're not getting cited by ChatGPT in those queries. Most marketers I work with have never opened Bing Webmaster Tools. (Don't sneer. I hadn't either until this year.)
The ranking layer does its own re-rank. Pages with clear, structured answers near the top get pulled. Pages with original first-hand data get pulled. Pages that look like AI-generated content from the same model often get filtered out, which is its own particular irony.
Claude
Claude uses its own retrieval. Anthropic hasn't published the full mechanics, and I'm not going to pretend I know them when I don't. What I can tell you from running the Claude browser extension on my own site for an audit is that Claude is unusually good at picking up first-hand experience signals. It cited my own posts back to me in the audit, recognised me as the author, and ranked my own content above generic SEO advice posts when summarising my niche. That's an entity recognition layer working harder than I'd expected.
Perplexity
Perplexity's whole product is "answer with citations." It's the easiest of the four to optimise for, in my experience, because it's the most transparent about what it pulled and why. Run a Perplexity query in your niche and watch which sources it cites. Patterns will emerge fast.
The pattern I see most. Perplexity loves recently-updated, well-structured content with clear question-answer formatting. It loves Reddit threads. It loves YouTube transcripts. It cites them constantly. If your topic doesn't have a Reddit thread or a YouTube video for the model to triangulate against, your post on its own won't survive the ranking step.
Google AI Overviews
Google AI Overviews pull from Google's own index. The signals overlap heavily with regular Google ranking, with one big exception. AI Overviews lean on FAQ schema, structured data, and clean question-formatted H2s much harder than the standard organic results.
If you've ever wondered why some posts get pulled into AI Overviews and others don't, the answer is usually structure. The post that wins has the question in an H2, the answer in plain prose immediately after, no hedging, no preamble, no "let's first define our terms." The model wants the answer. Give it the answer.
What this means in practice
The four engines have different mechanics but they overlap in three places. They all reward extractable answers. They all reward original information you can demonstrate is yours. They all reward distributed authority across multiple sources.
If you build content that hits those three, you're not picking which engine to optimise for. You're showing up across all of them at once.
The takeaway. AI engines have different retrieval layers but converging ranking signals. Build for extraction, originality, and distributed authority and you cover all four.
How do you write content that gets cited by AI engines?
Six things. In order of importance, based on what I'm doing on my own site this year.
Put the answer in the first 100 words
Old SEO advice was "build up to the answer." New SEO advice is the opposite. Lead with the answer. Two to four sentences, plain prose, no preamble. The model is scanning for the most extractable response in the post. If the answer to "what is X" doesn't show up until paragraph six, the model often moves on.
I write the answer first, in the post itself. Then the hook above it. Then the rest. It's a backwards way of writing and it works.
Use question-form H2s
If a real human might Google the question, that question goes in an H2. Not a clever rephrasing. Not a marketing-team title. The plain question.
"How do I lower my cost per lead?" beats "Driving Down CPL." "What is AI SEO?" beats "Demystifying AI Search." It's not glamorous. It works.
Underneath each question H2, the first paragraph is a definition-style answer. Front-loaded. Plain prose. Then the supporting detail.
Name specific entities, dates, and numbers
Generic content does not get cited. Specific content does.
"Most marketers" doesn't get pulled. "Roughly 47% of B2B marketers in a 2024 HubSpot report" does. "A major AI tool" doesn't get pulled. "Claude Opus 4.7" does. "Recently" doesn't get pulled. "21 April 2026" does.
Every paragraph that could contain a specific name, date, tool, or number, should. This isn't padding. It's the entity-richness layer. AI engines build a knowledge graph from named things. Generic prose has no nodes for the graph to attach to.
First-person experience signals
The biggest E-E-A-T uplift in the last year, for me, has been writing about real work I've done with real numbers. Not "here's how to use AI for lead generation." Here's what happened when I ran AI for lead generation across nineteen client accounts in the last six months.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Models can tell the difference. Or rather, the patterns associated with first-hand reporting are different from the patterns associated with summary content, and the engines are increasingly weighting first-hand reporting harder.
If you've done the work, write about the work. If you haven't done the work, do the work, then write about it. There isn't a shortcut.
Build internal topical clusters
A single brilliant post on a topic isn't enough. The AI engines look at your site as a whole and decide whether you're a serious source on a topic or a one-hit wonder.
If you want to be cited on AI for marketing, you need ten or fifteen connected posts on AI for marketing. Each one linking to the others with clean, descriptive anchor text. The cluster is the unit, not the post.
I run six or seven of these on my site. The AI cluster has a piece on building a one-person AI business, the AI for lead generation system I run, a rant about how to stop ChatGPT sounding like a corporate robot, and a few more. They link to each other constantly. The cluster ranks together.
Distributed authority across the web
The bit most people skip. Your post on its own won't carry an AI SEO win. Your post plus a podcast appearance plus a guest spot plus three LinkedIn posts plus a YouTube video plus a Reddit thread will.
The engines triangulate. They want to see you said the same thing, in your own words, in five different places online. Then they trust you.
This is why showing up on LinkedIn consistently in your niche has more SEO value in 2026 than it had in 2020. Not because LinkedIn ranks you. Because LinkedIn corroborates you.
The takeaway. Extractable answers, named entities, first-person evidence, topical clusters, and distributed authority. Do all five and the AI engines stop having a choice about whether to cite you.
How do you optimise for Google AI Overviews specifically?
Four things. They overlap heavily with the wider AI SEO list, but Google AI Overviews has its own quirks.
FAQ schema and clean question H2s
Google AI Overviews pulls hard from FAQ schema. If you've got a strong FAQ section at the bottom of your post with proper schema markup, your odds of being pulled into the Overview block on the equivalent search go up sharply. I add an FAQ section to every pillar post on my site for exactly this reason.
The FAQ section doesn't have to be long. Five to eight questions. Each answer 40 to 100 words. Plain prose. No marketing fluff.
Updated content
Google AI Overviews disproportionately pulls from recently updated content. A 2019 post on a fast-moving topic almost never gets pulled. The same post, updated in 2026 with current data and a new last-modified date, often does.
This is part of why the content rescue operation I ran on my own site involved updating 47 of my best-performing posts before publishing anything new. Updates compound. New content from a low-authority site is always going to lose to updated content from a higher-authority site.
Structured data beyond FAQ
Article schema. Author schema with sameAs links to your social profiles. HowTo schema where it fits. Breadcrumb schema. The technical bits a lot of business owners outsource and never check.
I went through my entire site's structured data in the rebuild this autumn. Found three different schema types competing on the same pages. Cleared the lot, set one canonical schema strategy, and the impressions in Google Search Console climbed within a fortnight.
The de-indexing point
This is the one nobody talks about. If your site has hundreds of low-quality pages dragging your overall domain authority down, your good pages won't get pulled into AI Overviews regardless of how well-optimised they are. Google judges the whole domain.
I de-indexed roughly 1,300 pages on my own site this autumn. Old course pages, dead category archives, a graveyard of paginated tag pages, all sorts of stuff that was indexed but should never have been. The site felt smaller afterwards. The rankings went up.
If your site has more than a couple of hundred indexed pages and you're not sure why, that's where to start. Open Search Console, look at the Pages report, and audit. The pages your good content is competing against most aren't your competitors' pages. They're your own bad pages.
The takeaway. Google AI Overviews rewards structured data, recency, and clean domains. Check your schema, update your good posts, and remove the rubbish that shouldn't have been indexed in the first place.
What most people get wrong about AI SEO
Three big mistakes. I see all three in client audits, all the time.
Treating it as a tooling problem
The first mistake is buying an AI SEO tool, plugging it in, and waiting for the rankings to come. There isn't a tool that solves AI SEO for you. There's a system. The tools fit inside it.
I use a small handful in my own work. ChatGPT, Claude, Perplexity, and Search Console are the four I open every day. Beyond that, the tools that promise to "rank you in ChatGPT in 14 days" are mostly selling you a checklist someone else's content team already had.
The system is the thing. The tool just helps you execute the system faster.
Writing for the model instead of the reader
The second mistake is over-optimising for extraction at the expense of being readable. You can write a post that's nothing but bulleted Q&A pairs and it'll get cited. It'll also be unreadable, nobody will share it, and the brand-building benefit of being cited disappears because the post itself is a robot.
The version of AI SEO that wins is content a human would forward to a colleague AND a model can summarise cleanly. Both. Not either-or.
Skipping the foundations
The third mistake is jumping straight to "how do I rank in ChatGPT" without checking whether your site is technically sound, whether your content is updated, whether your internal linking makes sense, whether you've got a sensible site architecture, whether your schema is clean.
No structural AI SEO trick is going to compensate for a broken foundation. I see business owners chase the new thing every quarter and never sit down to do the boring work. The boring work is the win.
(I am as guilty of this as anyone. The reason my rebuild this autumn even happened is that I'd skipped the boring work for three years and the chickens came home.)
The takeaway. AI SEO failures are usually system failures, voice failures, or foundation failures. Tools won't save you. Robot prose won't save you. Skipping site health won't save you.
What nine weeks of AI SEO work on my own site looks like
The reason I'm writing this post with any conviction is that I've spent nine weeks running it on my own site and writing about every step. The series is public. Numbers are real. The bits that didn't work are in there too.
Quick tour.
Week 1 was the de-indexing. Roughly 1,300 pages removed from Google's index. Old course landing pages, broken category archives, paginated tag pages, the lot. The site dropped from around 2,000 indexed URLs to under 700. Three weeks later, average position for my target keywords had moved up by twelve places. The site was smaller and louder.
Week 4 was the content rescue operation. Forty-seven posts on my site were performing decently but had data from 2022 or 2023, broken images, and weak internal linking. I went through every one of them, updated the data, fixed the structural issues, added FAQs where they were missing, and re-published. Traffic on those forty-seven posts roughly doubled in six weeks. Cost, most of two weeks of my time. Outcome, the highest-trafficked content cluster on the site is now the rescued cluster, not the new content.
Week 8 was the Claude browser extension audit. I installed the Claude browser extension and let it loose on my site for an SEO audit. It found seventeen issues across structured data, internal linking, and topical coverage that my regular SEO crawler had missed. Two of them mattered (a canonical tag pointing to a redirect chain, and a duplicate H1 issue across the AI cluster). I fixed them in an afternoon. Traffic on the AI cluster moved up the following week.
Week 9 was the newsletter design experiment, which sounds off-topic but isn't. The newsletter feeds the AI SEO loop because every issue mentions the recent posts, every issue gets archived as a public page, and every issue produces signals the AI engines pick up. The redesigned newsletter has a 70% open rate on a 15,000-subscriber list. That's distributed authority showing up in actual numbers.
What I learned across the nine weeks. The work that moved AI SEO most was the de-indexing and the content rescue. The structural changes, the schema cleanup, the new content, all of those mattered, but they mattered second. The single highest-impact week of the rebuild was the one where I deleted things.
That's not the answer anyone selling AI SEO services wants you to hear. It's the one I keep seeing in client work too.
The takeaway. The AI SEO playbook that worked for me on my own site this autumn was 60% removing rubbish, 30% updating good content, and 10% writing new. Reverse the proportions and you'll be busy without being found.
AI SEO FAQ
How long does AI SEO take to show results?
Faster than regular Google SEO, in my experience. Google ranking moves take three to six months from the work to the visible result. AI SEO citations can change in two to three weeks because the retrieval layer updates more frequently than the Google index. Within a month of the de-indexing on my site, I was being cited in Perplexity on queries where I'd been invisible the month before.
Do I need a separate AI SEO strategy if I already do SEO?
Not separate. Layered. Your existing SEO foundations should be the base. The AI SEO additions sit on top. Extractable answer formats, question-form H2s, FAQ schema, entity richness, distributed authority across the web. Treat AI SEO as the next layer of the same job, not a replacement for what you've already built.
Which AI engine should I optimise for first?
Whichever your audience is using. For B2B owners and marketers, Perplexity and ChatGPT are where the queries with buying intent live. For broader consumer queries, Google AI Overviews still has the largest reach. For developers, Claude. Run a few real queries in your niche across all four and watch which ones surface useful answers. That tells you where your audience is.
Does AI SEO replace traditional Google SEO?
No. It expands it. Google still drives a meaningful share of search traffic and will continue to for a long time. The AI engines drive an additional and growing share. You want to be found across both. The good news is that 80% of the work overlaps. Do the foundations once, structure for extraction, and you cover both at the same time.
How important is FAQ schema for AI SEO?
Important for Google AI Overviews. Less important for ChatGPT, Claude, and Perplexity, which don't lean as hard on schema for retrieval. That said, FAQ sections themselves (whether or not you mark them up) help all four engines because the question-answer format is the most extractable structure they can find. I include an FAQ on every pillar post for exactly this reason.
Can I use AI to write my AI SEO content?
You can use AI in the workflow, but the model can't write the post for you and have it work. The reason is that AI engines have started filtering out content that looks like it was written by the same model. Use AI for outlines, research, fact-checks, and polish. Write the prose yourself, in your voice, with your first-hand information. That's the bit that gets cited.
How do I track whether AI SEO is working?
Three things. Search Console for the Google AI Overviews traffic. Direct query checks in ChatGPT, Claude, and Perplexity for whether you're being cited on your target queries (do this weekly). Branded search volume, because if AI engines are recommending you, branded searches go up. None of the standard SEO tools track this well yet. The manual check is the most reliable signal.
Final word
The version of AI SEO that's being sold this year is mostly a checklist. The version that works is mostly a system. The system is your foundations, your topical clusters, your structured answers, and your distributed authority all working together, with the AI engines as one of many places that pulls from the work.
I've been doing this kind of work for over twenty years. The tools change. The platforms change. The job is the same. Build something a real audience finds useful, structure it so it's easy to find, prove you're the person to listen to with first-hand evidence, and keep showing up.
If anything, AI SEO is more honest than the SEO that came before it. The shortcuts don't work. The link farms don't work. The thin content doesn't work. The model has read all of it before. It can tell.
What works is the unglamorous version. Update your old content. Delete the rubbish. Write what you know. Put the answer first. Name the things. Show up in five places, not one.
That's the whole post in a paragraph. The rest of it is just the receipts.
If you're sitting on a site that hasn't been touched in two years and wondering why nothing's moving, this is your nudge. Open Search Console tonight. Look at the Pages report. Find the rubbish. Start there.
The work is dull. The compounding is not.
Stay in touch
Every Sunday morning I send a newsletter to fifteen thousand entrepreneurs, marketers, agencies and business owners. It's where I write about the experiments before they make it to the blog. Sign up here.
If you want help building this kind of AI SEO work into your own business, that's what I do as a consultant and coach. I work with a small number of business owners and in-house marketing teams at any one time, helping them rebuild around AI tools so they stop paying for things they can do in-house in a fraction of the time. The AI SEO work above is one of the workflows I run with clients. Details on the Work With Me page here.
