AEO Guide: Optimize for AI Search — Step-by-Step Guide & Transcript

[00:00] they just open ChatGPT and ask for the answer AI lists out the key points for you one by one most people leave after reading the answer

[00:14] ChatGPT's weekly active users have hit 900 million and found that in Google search results the click-through rate of the #1 page drops 58%

[00:28] your traffic can still be stopped at the door by the AI's answer has grown nearly 10x in a year averaging just 0.25% of a site's total traffic

[00:41] by the time AEO becomes mainstream when AI answers a question Answer Engine Optimization

[00:55] Optimization around since the featured-snippet era whichever you hear, it's the same

[01:07] retrieve paraphrase correctly is worse than not being mentioned

[01:20] AEO is not a separate-from-SEO you can directly crank up you don't need any special AI markup

[01:33] expert content lay the foundation well covering the five steps of SEO

[01:45] I'll explain it in five steps what data each AI uses how to make AI recognize your brand

[02:00] how AI produces answers what happens in between? there are several layers in between

[02:12] it rewrites your question one question fires off third-party research measured

[02:25] expands into about 10 sub-queries are phrasings no one has ever actually searched you can no longer optimize for a single keyword

[02:38] your content needs to cover the entire topic it recalls a batch of candidate documents then feeds those documents to the model to read

[02:50] finally the model reads these documents here's the key it might not get recalled at the search stage

[03:02] or it might make it into the model's context so AEO breaks into two problems to tackle second, generation visibility

[03:14] academic research even splits citation into two stages and the source's content actually being absorbed into the answer doesn't mean AI actually adopted your view

[03:28] if AI can't find you at all but if AI finds you and doesn't cite you among Google AI Overviews citations

[03:41] has already dropped from 76% in mid-2025 ranking well still has an edge SEO is the foundation

[03:53] what data does each AI use? ChatGPT has its own web search Google's AI Overviews and Gemini

[04:06] Claude also has its own search bot the top 50 most-cited domains in other words

[04:20] you can't focus on just one platform one company usually has more than one crawler OAI-SearchBot is the search crawler

[04:32] GPTBot is the training crawler ClaudeBot for search-index quality ClaudeBot for model training

[04:47] you can allow AI to search and cite your site the way to do it is in robots.txt that tells crawlers where they can and can't go

[05:01] ChatGPT and Claude can find you when answering questions first, a wrong setting causes friendly fire you vanish straight out of AI's answers

[05:15] second, Google's situation is special there's a separate control called Google-Extended it can use your content as reference material

[05:29] search and AI disappear together major AI companies promise to obey it won't disappear from training data because of it

[05:43] one study scanned the robots.txt of 140 million sites many of them from copying a template or keeping a default setting spending a minute to open your own robots.txt and check is well worth it

[05:58] important content must exist directly in the HTML one preferred URL per concept this JavaScript point is stricter for AEO than for SEO

[06:13] but doesn't execute them what it gets is an empty shell after you update content

[06:25] to proactively tell search engines like Bing to re-crawl you may have heard of something called llms.txt honestly, so far

[06:39] Google's people have also said publicly but don't put it ahead of the earlier stuff last episode we used Claude SEO to scan technical SEO

[06:52] it's a free open-source command-line tool it scores across 8 categories whether your robots.txt rules have issues

[07:07] and finally lists a prioritized fix checklist GEO research a high score doesn't mean AI will definitely cite you

[07:21] llms.txt Part three first picture the scenario where AI cites content

[07:34] and lines it up next to passages from other sites to compare and be understood on its own when

[07:46] "it works really well" AI has no idea what "it" is answer-first

[07:59] then expand on conditions the whole page can be designed in three layers trial dates and limits

[08:11] the conversion layer this structure serves two kinds of readers at once is the fix for "read the answer and leave"

[08:25] is put things on your site that require visiting to use transaction features length doesn't equal quality

[08:40] the correlation between word count and being cited was only 0.04 over half of cited pages were under 1,000 words the third principle, also the most defensible one

[08:55] there's a study published at KDD 2024 called GEO after adding reliable citations and statistics to content visibility in generative engines rose by up to 40%

[09:08] not that adding numbers guarantees a 40% jump is more likely to be picked by AI than vague rewrites

[09:20] clear calculation methods when AI wants to cite real test results don't use AI to mass-produce templated Q&A pages

[09:33] mass-produced content lacking user value being cited isn't the same as being paraphrased correctly and found only about half the sentences in the answers

[09:48] prices to lower the chance AI paraphrases them wrong check whether it's getting your stuff wrong

[10:01] they made up a brand out of thin air then took questions to each AI stated the fake data as fact

[10:15] because in 84% of responses between a vague truth and a concrete fabrication to spell out every basic question about your brand

[10:30] concrete enough that AI has no blanks for others to fill when AI organizes data it faces a hard problem another way on social

[10:44] and the signals you've built up get diluted on the website, in structured data use JSON-LD structured data

[10:59] this is a block placed in the page source one thing to add here and the content your schema states

[11:11] make freshness real it doesn't just look at the date on your page before it re-adopts it

[11:23] changeable info like prices, rules, versions research compared is on average 25.7% fresher than traditional search results

[11:37] the first three steps are all on your own site comes largely from how others write about you the number of times a brand is mentioned online

[11:51] higher than traditional SEO metrics like backlinks and domain rating brand mentions on YouTube and ChatGPT visibility

[12:03] the strongest of all factors but the direction is clear is mentioned in a consistent way

[12:15] you also need to cultivate an off-site presence your position in listicles and comparison articles Part five

[12:28] what about AEO? compile 50 to 200 branded, unbranded, comparison-type, and recommendation-type all included

[12:42] then regularly ask these questions to ChatGPT, Google AI, Copilot, Claude was the content paraphrased correctly the same question on different days

[12:56] test repeatedly and look at the ratio has an AI Performance report that directly shows how often your site is cited by Copilot and Bing AI

[13:09] Google Search Console is also starting to provide AI-search performance data carry a tag on the URL group traffic from each AI separately

[13:23] AI sometimes hallucinates sending users into a 404 is 2.87x that of Google Search

[13:36] find these 404 URLs that have traffic in your analytics tool and you catch visitors you would have lost but whatever tool you use

[13:51] is what it says correct? the real substance most of it is an extension of last episode

[14:04] to manage search crawlers and training crawlers separately write important passages in a form that can be cited on their own

[14:16] cultivate off-site brand mentions fourth regularly measure your visibility and accuracy in AI answers

[14:29] validating schema what wording the brand should use you might wonder what happens if you don't do all this

[14:43] whenever a search result shows an AI summary it's already happening users will ask it

[14:56] AI stitches together whatever it finds online it's having your AEO ghostwritten by someone else spend five minutes

[15:10] this is the most common take ten questions your customers would ask is it saying the right thing

[15:22] change the opening to answer-first you can finish it in an afternoon those are things to keep doing each month going forward

[15:35] but the logic behind the entry point hasn't is who gets recommended if you liked today's content

[15:47] bye