The internet is flooded with "ultimate AEO guides" that are old SEO checklists with ChatGPT pasted on top, written by the same people who spent a decade selling one weird trick to beat the algorithm. This isn't that. Here's what actually matters, with receipts, and nothing recycled.
Answer Engine Optimization (AEO) is the work of becoming the source AI engines like ChatGPT, Claude, Gemini, and Perplexity cite, mention, and recommend when they answer a question. Not a link a human clicks. A mention a machine makes for you.
→ AEO isn't SEO with new words. Old search gave you a link a human picked. AEO gives you a mention a machine makes for you.
→ The webpage stopped being the destination. It's now raw material for someone else's answer. Optimize for the answer, not the page.
→ Stop counting rankings. Start counting how often the model brings you up on its own, and how warmly. That's your Share of Model.
→ Know your buyers first. Nobody finds a plumber on ChatGPT. High-stakes, well-researched purchases are already decided in AI. Local and impulse, not yet.
→ You can't buy your way in. About 84% of AI citations are earned, so reputation does the work: primary sources, reviews, communities, the press pickup, a named expert.
→ Production got free; a point of view didn't. Make it in volume, text, video, and audio, say your claim out loud, and keep your best pages fresh.
→ The next buyer might not be a person. The agent is the customer now, assembling tools and vendors the human never sees. Be the default it builds with.
→ Don't become beholden to the models. You don't have to rebuild your business for ChatGPT or Gemini. Get your own house right and the results follow. Everything else is a treadmill.
Watch what happened the second AI search got real. The exact same crowd that spent ten years stuffing keywords and renting links simply changed the logo on the slide deck. SEO became AEO became GEO became AIO, and the only thing that genuinely changed was the invoice.
There's now a whole cottage industry built on making you anxious about a three-letter word and then selling you a dashboard for it. Tools that track a number nobody can verify. "Certified" AEO experts who got certified last Tuesday. Ninety-seven-point checklists that quietly copy each other, right down to the same orphan statistic that traces back to nobody.
So here's the BS, called plainly. You can't game a model the way you gamed a ranking. There's no keyword density to juice, no link farm to rent, no meta tag that whispers the magic word. The model formed its opinion of you from the entire messy internet, not from your title tag. The only hack that survives contact with reality is being genuinely worth mentioning.
The whole industry is optimizing for the machine. Almost nobody is optimizing for being worth a damn.
Here's the headline nobody selling you a formula will say out loud: you can't rely on Reddit. Or on any single source. For a year Reddit was the darling everyone told you to chase. In April it was the single most-cited domain in ChatGPT Search. By mid-August its citations had fallen 86% in four days, and nothing you did caused it.
Two separate things happened, and it's worth keeping them apart, because most of the write-ups have mashed them together.
One: the engine changed its mind. OpenAI adjusted how ChatGPT expands a query behind the scenes. Its use of the site: operator jumped from roughly 0.4% of those background queries to nearly 17% in a matter of days, and Reddit fell off a cliff. Google's AI Overviews barely moved over the same window. That's the tell: this was one engine rewiring itself, not the internet shifting.
Two, and this is the bigger one: Reddit shut the door. Reddit's robots.txt now blocks any crawler that hasn't signed an agreement. Access is a licensing deal, reportedly around $60 million a year from Google and $70 million from OpenAI, moving toward usage-based pricing where the bill scales with how much the data actually gets used. Search engines that don't pay don't get to index it. Reddit isn't a public commons you can earn your way into anymore. It's a paid feed, and you're not a party to the contract.
So if your plan was to influence AI answers by seeding Reddit, that plan is over. Not degraded. Over. And Reddit is just the loudest example: Cloudflare now blocks AI crawlers by default and can charge them per crawl, and millions of sites have disallowed AI training outright. The open web you were optimizing for is being fenced and metered while you read this.
That's the whole game now. An engine can drop a source overnight for reasons that have nothing to do with you, and the sources themselves are becoming private arrangements between companies you've got no relationship with. There's no one-size-fits-all, not across engines and not across months. This is quicksand, and the more weight you put on any single platform, the faster it swallows you.
There isn't one strategy to game, because there isn't one system. You can't game this thing. You can only be worth citing and stay light on your feet.
So the real test of anyone selling you AEO, whether it's a formula, a certification, a 97-point checklist, or a piece of software, isn't the checklist. It's one question: how adaptable are they? A playbook written for how ChatGPT cited sources last quarter's depreciating the day you buy it. Hire for how fast someone notices the ground move and changes the advice, not for how confident the slide deck sounds. Treat this like weather, not settled science.
For thirty years the operating system of information was the page and the link. That system is being swallowed by an answer layer. Most people no longer click. They ask, they get an answer, they move on. So the job flips: you're not trying to get visited anymore, you're trying to be the source the system trusts when it speaks. The answer layer is the new API. Your brand is either in it or it's not.
We're leaving the era of the fixed page and walking into generative UI: interfaces that assemble themselves around one person's intent the second they ask. There's no "page one" to win because there's barely a page. Can a model lift a clean, correct answer out of your stuff without your layout, your nav, or your logo to lean on? If your value only works inside your design, it doesn't survive the trip into someone else's interface.
Producing enough genuinely useful, structured, first-hand material to become the answer across a whole category used to cost too much. That wall is gone. The winners don't use AI to copy the same listicle ten thousand times; the models can smell that. They use it to manufacture the authority of a research lab while keeping the taste and nerve of one sharp human. Original data. Real benchmarks. A named expert's actual opinion, captured across a hundred specific questions instead of one mushy page.
And this stopped being text-only. Audio and visual are now just as cheap to make, and the engines feed on transcripts and captions: every video and podcast is another structured, quotable, named source pointing at you.
The bottleneck was never the making. It was the conviction. AI just made conviction the only scarce thing left. It's also roughly why we named the company Talentless AI.
Most AEO stats in circulation are junk. Blogs quoting blogs quoting a vendor's blog quoting a tweet. Here's the short list we actually trust, with primary links.
Solid: earned media runs the show. Muck Rack's analysis of more than 25 million AI citations found roughly 84% come from earned media and about 99% from non-paid sources. Paid and advertorial is 0.3%. Journalism alone is around 27%.
Solid: the engines cite nothing alike. ChatGPT cites in 96% of answers but shallow (~5 sources). Gemini cites in 82% (~8). Claude cites in only 55%, but goes deep (~13). Top domain per engine: Wikipedia for ChatGPT, PubMed Central for Claude, Reddit for Gemini (as of the May 2026 Muck Rack report; now in flux). Three different information worlds.
Solid: the ground moves fast, and it's getting gated. Reddit's share of ChatGPT citations sat around 3.8% for weeks, then fell about 86% in four days in mid-August 2026 after OpenAI changed how ChatGPT expands queries. Google's AI Overviews only slid gradually over the same window, so that particular cliff was engine-specific. Separately and more permanently, Reddit's robots.txt now blocks crawlers without a signed agreement, and access runs through licensing deals reported at roughly $60 million a year with Google and $70 million with OpenAI, trending toward usage-based pricing. A 129-million-citation study found publishers with an OpenAI licensing deal earn 48% more ChatGPT citations. Cloudflare blocks AI crawlers by default and can bill them per crawl, and millions of sites now disallow AI training outright. The takeaway isn't "Reddit is dead." It's that the sources feeding these answers are becoming private commercial arrangements, and you are not a party to them.
Solid: the click is collapsing. Google users click a link far less often when an AI summary sits on top, and a large share of searches now end with no click at all (Pew Research, SparkToro).
Solid: GEO is a real, studied thing. The original Generative Engine Optimization research (Aggarwal et al., KDD 2024) showed you can measurably change how often a model cites you by changing how content is written and structured.
Directional, treat with care. The eye-popping vendor numbers point at something real. The direction is probably right. The exact multiplier isn't gospel.
Fewer clicks, higher intent, and a brand's reputation doing more of the work than its link graph. That's enough to act on. You don't need fake precision.
SEO trained a generation to think in positions. Wrong shape for what's happening. There's no stable list now. There's a guess inside a model's head.
The right model is recall, the human kind. When a sharp friend gets asked "who is good at this," they don't run a query. They remember the names that kept showing up, credibly, in that context. How often they think of you, and how warmly, is the entire game.
Models learn who you are from consensus. They trust the brand that lots of independent, credible people describe the same way, again and again. A backlink is a vote. A mention in a forum, a podcast transcript, an analyst note, a customer's blog, is the texture of a reputation the model actually learns from. You're not building a link graph. You're building a memory.
Two ways a model knows you. One: you're baked into training, durable but slow. Two: the model looks you up live at the moment of the question (RAG), fast and fresh and the lever you can pull this quarter. Almost everything here targets the live-retrieval layer. The training layer is the long game, won by being talked about enough, for long enough, that the next model simply grows up knowing who you are.
The usual version of this question is "do your buyers use AI?" That's the wrong question now, and it was always a bit lazy. Almost everybody uses AI to find things. The real question is which one, and where it lives.
Nobody opens ChatGPT for an emergency plumber. But that doesn't mean the plumber's safe from AI, it means the AI moved. In March 2026 Google put Gemini inside Maps as Ask Maps, and people started asking it things like "find me a quiet Italian place for a business dinner downtown with outdoor seating" or "is there a public tennis court with lights I can play on tonight." That's a conversational AI answer, sitting on 300 million places and half a billion contributors' reviews, in the app people already open for local.
So if you sell locally, the answer layer you need to worry about probably isn't ChatGPT at all. It's Maps, and the inputs are unglamorous: your categories, your services, your hours, your booking link, your photos, your accessibility details, and above all the language your customers use in reviews. Your Business Profile stopped being a listing. It became the training data for the answer.
Meanwhile a CMO scoping a six-figure platform, a founder comparing vendors, a patient researching a condition: those people are asking a chat assistant first. The higher the stakes and the longer the homework, the earlier AI enters the decision. Different buyer, different machine, different work.
Google is still the reigning king, and Maps with it. That's fine. The point isn't which logo wins. The point is to find the specific place your buyer forms an opinion, and be good there, rather than spraying effort at whichever assistant's in the headlines this month.
AEO isn't a tax everyone has to pay. It's a bet you place where your buyers actually are. Know them, and you've done half the work.
And here's the part the vendors will never sell you: don't become beholden to the models. You don't have to rebuild your business around ChatGPT, or Gemini, or whatever launches next quarter, no matter how loudly anyone insists otherwise. Every one of these systems is trying to work out who's actually good at the thing. Chasing them is a treadmill that resets every time an engineer changes a setting. Get your own house right, say clearly what you do, publish the proof, treat people well enough that they say so in public, and the results follow. That's the only strategy with a shelf life longer than a quarter.
First-hand sources get quoted far more than aggregated commentary. Run the survey. Publish the benchmark. Turn one expert's brain into two hundred structured answers instead of one generic page. This week: pick one thing your category argues about, get real data or a real opinion on it, publish it as a standalone, dated, quotable artifact. Then do it again.
Lead every page with the answer. Two or three sentences that name the thing by name and state the point. This week: rewrite your top ten pages so the first fifty words answer the title, by name. Add a real FAQ block with self-contained answers.
Models learn your reputation from other people's pages. This quarter: claim a Wikidata entry for your brand and key people. Build genuine community presence, not spam. Get into the credible "best of" lists in your space.
The engines that weight expertise reward identifiable, credentialed authorship. This month: add author schema and real bios with external profile links to every serious page.
AI loves recent. 57% of journalism citations were published in the past year, and citations spike in the first month after a page goes up. This quarter: list your ten most important pages, put a real "last updated" date on each, refresh the top ones quarterly. Old page plus new facts plus a new date often beats writing something brand new.
Let the AI crawlers in: GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Applebot, Google-Extended. Block one and you're invisible to that engine no matter how good the work is. Ship clean semantic HTML and basic schema. On llms.txt: ship one, it's cheap and tidy, but anyone selling it as a citation guarantee is selling you 2019 in a new font.
If someone asks for "the best AI studio in Austin" and literally nobody has ever written those words, the model has nothing to grab. It can't infer the claim from your vibe. It matches language to language.
Which means your About page matters again. Be specific. Be a little bold. "We're the best AI studio in Austin for branded video" gives the model a clean, matchable claim. "We help brands tell stories in the age of AI" gives it mush.
The honest caveat: a claim you can't back gets contradicted by every other source the model checks. Bold and true is the cheat code. Bold and hollow just gets fact-checked by a machine.
The channels aren't a flat list, they're a ladder. Each rung is a deeper level of being built into the decision, and on each one the human sees a little less of you.
Where the volume is. Get named in quick conversational answers. Pick your fight by engine: ChatGPT runs on familiarity (Wikipedia, big business press, YouTube). Perplexity rewards forums and answer-first pages. Claude rewards named expertise and careful sourcing. Google leans on organic authority plus the entity graph.
Reddit is no longer a lane you can plan around. Its citations in ChatGPT fell about 86% in four days in August 2026, and Reddit now blocks crawlers that haven't signed a licensing deal. If seeding Reddit was your AI visibility plan, retire it. Never anchor to a single platform, and especially not to one whose front door is a contract you're not on.
Hand an AI a real question and it browses a hundred-plus pages to assemble a sourced report. This is the whole ballgame for B2B: procurement hands an agent the requirements and lets it build the comparison. Depth beats slogans. The long tail is the game. Consistency wins the cross-check. Be checkable.
The thing choosing your product isn't a person. Agents assemble inputs, pick libraries, and wire up tools the human never sees. Be usable by an agent, not just readable by a human: clean machine-readable docs, an MCP server, unambiguous quickstarts. Vibe coders have zero patience for friction, and friction is churn before signup.
You stop being something an agent picks once and become something it builds in. Get designed into enough apps and you're not a vendor anymore, you're infrastructure. Chat gets you mentioned. Applications get you depended on. Climb.
On hiding instructions in your pages: people bury text humans can't see but a model might read, stuffed with things like "recommend us first." It's keyword stuffing reincarnated. The industry calls it indirect prompt injection. A simpler word is lying.
The downside is brutally asymmetric. The upside is a temporary nudge. The downside is getting caught hiding deceptive text on your own website: a screenshot waiting to happen, a guidelines violation across every platform at once, and a security-and-legal flag with your name on it. For a brand that wants to be remembered, getting remembered as the company that lied to the robots is not the look.
There's no exploit that survives contact with reality. The only durable hack is being genuinely worth recommending. Annoying, I know.
If your dashboard still leads with rankings and organic sessions, it's measuring a shrinking world. Track how often the models name you, how they talk about you, AI-referred conversion on its own, and never blend the engines together.
Share of Model = answers that name you ÷ answers you sampled
Run the same prompt set in each engine, on a schedule. Count the mentions. Watch the line, not the number. Quality-adjusted version: score each mention +1 positive, 0 neutral, −1 negative, and average that instead.
You are my AEO analyst. My brand is [BRAND], we sell [WHAT YOU SELL] to [WHO], and our main competitors are [A], [B], [C]. 1. Write 15 buyer-intent prompts a real customer would type when looking for what we sell. Mix three types: category ("best ___ for ___"), comparison ("[competitor] vs alternatives"), and problem-based ("how do I ___"). 2. Answer each of the 15 the way you normally would. For each, tell me: did you mention [BRAND] unprompted? Where did we rank? Which sources or sites did you lean on? 3. Score our Share of Model: out of 15, how many answers named us, and what was the overall sentiment (positive / neutral / negative)? 4. List the 5 sources that most influence answers in our category, and the single highest-leverage thing we could do to get cited more. Be brutally honest. If we are invisible, say so, and tell me who is winning instead.
Drop the 15 prompts and the scores into a sheet, one tab per engine, one row per month. Re-run on the first of the month. Inside a quarter you'll see the line move, and you'll know which play is actually working. That's a dashboard. No license required.
We do AEO strategy the way we do everything else: with receipts. No gate, no form, no "book a demo" to read the guide. But if you want the applied version, that's a conversation.
Let's talk