This is the methodology itself, not the sales pitch for it. Every engagement I run - Sprint or Retainer - is an application of these same four steps, in this order. Skipping a step, or doing them out of sequence, is the most common reason AEO work doesn't produce citations.
01. Define the pillars
What you do, who it's for, and the three or four claims every AI answer needs to reinforce about your firm. Not five. Not a dozen. Models selecting sources need clear, specific, repeatable claims they can confidently cite - vague messaging like "full-service" or "results-driven" gives them nothing to extract. This step happens before a single word gets written, because every piece of content afterwards has to reinforce the same pillars. Change the pillars halfway through and you reset the signal instead of building it.
02. Structure for extraction
Content engineered for how LLMs actually parse and cite a page - not for how search engines rank one. In practice that means direct answers positioned prominently near the top, clear heading structure, and plain, readable HTML. Case studies locked in PDFs, results buried in JavaScript-rendered pages, and walls of unstructured text are the most common ways otherwise-strong content stays invisible to a model. I've written in more detail on why this specific step trips up most agencies. I don't rely on FAQ schema markup to do this - Ahrefs tested 1,885 pages that added it and found the code itself made no measurable difference. What a model actually reads is the visible question-and-answer text on the page, so that's what I write for.
03. Publish at volume
Consistent, scheduled publishing so gains compound between articles instead of rising and falling with each one. Consistent volume matters more than any single article - visibility tends to fade without sustained output behind it. This is the step the case study demonstrates most directly: six articles, published on a defined cadence and all reinforcing the same pillars, generated 2,357 citations in a single week. Not sixty articles. Six, done consistently. See the full result → I schedule around a specific number, not a hunch: a citation's median half-life runs to about 4.5 weeks for content that exists in one place, roughly doubling when the same claim is also referenced elsewhere - which is exactly what consistent, connected publishing produces.
04. Track & redirect
Weekly citation tracking across eight major LLMs - ChatGPT, Perplexity, Claude, Gemini, Copilot, Grok, Meta AI and DeepSeek - feeding straight back into the following month's targets. I track all eight, not just the obvious ones, because a single-platform view is close to worthless: an analysis of 680 million citations found only around 11% of domains get picked up by both ChatGPT and Perplexity for the same query, so visibility on one platform says almost nothing about another. Citation count alone isn't the whole picture either: the fuller measure is whether you're visible on the specific prompts you targeted, how much AI-referred traffic actually reaches your site, and how the models describe you when they do cite you. Two firms can both get cited and come away with very different framing.
Same framework, different execution per platform
The four steps above hold regardless of which model I'm optimising for. How I execute steps two and four doesn't. ChatGPT, Claude and Gemini select sources in different enough ways that I run different tactics on each, not the same content plan three times over.
For ChatGPT, I confirm Bing Webmaster Tools is set up and build your own domain as the primary source - it leans on Bing's index and rewards owned content directly, and 60.9% of its citations trace back to a brand's own site. Claude gets closer to the opposite treatment: it barely searches the live web for most prompts, and when it does, the index behind it looks like Brave, not Google, so I check Brave visibility specifically and put the effort into earning press and review coverage instead - that's what Claude actually converts. For Gemini, I build entity signals - a Wikidata record, a Wikipedia mention where you genuinely qualify - because the link between a Google top-ten ranking and an AI Overview citation has weakened sharply in the past year, and a ranking alone isn't the guarantee it used to be.
None of this changes the four steps I run. It changes where step two's structuring effort and step four's tracking attention go, once the pillars are set.
Where this comes from
This isn't theory. It's the exact process behind the results in the case study - a UK digital marketing agency that went from zero AI search presence to citations across all 8 major LLMs in eight weeks. As a structured engagement, this framework is delivered either as an AEO Sprint for firms starting from zero, or an AEO Retainer for firms ready to sustain and expand it.
How this differs from SEO
SEO optimises for ranking algorithms and click-through. This framework optimises for extraction - how a language model reads, trusts and quotes a page when constructing an answer. The two overlap, but the content structure, the metrics, and the definition of "winning" are different. For more on the distinction: SEO and AEO aren't the same job, Is your SEO agency actually doing AEO? and why "more SEO" won't fix your AI search problem.