Your site is probably being crawled harder than ever right now — and getting less traffic for it. That is not a contradiction or a tracking bug. It is the new shape of the web: AI crawlers visit sites thousands of times to assemble answers, and those answers get read inside ChatGPT, Perplexity, Claude, or Google’s AI Overviews instead of on your page.

Answer engine optimization (AEO) is the discipline that grew out of that gap. It is not a rebrand of SEO. It asks a different question: not “how do we rank for this keyword,” but “when an AI writes the answer, does it name us — and does it say something good?”

Here is a practical playbook for getting cited, based on what is actually working for brands right now.

Stop reporting traffic as your headline metric

The first thing that breaks in AI search is your dashboard. Search Console impressions climb while clicks fall, and it looks like failure. It usually isn’t. More people are searching than ever; more of them finish reading inside the model.

If your monthly report still leads with organic sessions, you are measuring the one number guaranteed to decline. The metrics that matter now are citation frequency (how often you appear in AI answers), share of voice against named competitors, and sentiment — whether you are mentioned as an option or recommended as the answer.

This shift is uncomfortable for agencies especially, because sessions are what clients have been trained to expect. We covered how to reframe that conversation in what to tell clients when organic traffic drops, and the wider measurement gap in the metric most SEOs still aren’t measuring.

Prompts are not keywords — build a prompt index

A Google keyword averages three to four words. An AI prompt averages around 23. That difference is the whole game.

“Best running shoes” is a keyword. “I’m a runner with wide feet, I mostly run on broken city pavement, and my last pair gave me shin splints — what should I buy?” is a prompt. The model answers the second one with specifics, and whichever brand’s content addressed wide fits, urban surfaces, and shin splints is the brand that gets named.

You cannot chase prompts the way you chased keywords, because prompts are effectively infinite. What works instead is a prompt index: a curated, scored set of the questions that actually matter to your business — high commercial intent, realistic phrasing, mapped to where the buyer is in their decision.

Build it before you write anything. Score each prompt on two axes: how much revenue it touches, and how badly you currently perform on it. The intersection is your content roadmap. Our guide to what content needs to look like to earn AI citations covers the page-level execution.

A mention is not a recommendation

Brands celebrate the first time ChatGPT says their name. That is the floor, not the ceiling.

Think of AI mentions on a spectrum. At the bottom, you are listed among eight alternatives. In the middle, you are described accurately but neutrally. At the top — the position that actually moves pipeline — the model recommends you as the authority when someone asks who is best in your category.

Moving up that spectrum is a sentiment problem as much as a visibility problem. Models synthesize how the internet talks about you: reviews, comparison posts, forum threads, YouTube videos, podcast transcripts. That means brand reputation work and search work are now the same project. If there is bad material outranking you, deal with it directly — see how to remove negative content before AI answers cite it.

One channel consistently over-delivers here: YouTube is among the most-cited domains across every major LLM. If you have written off video as a brand-awareness play, that assumption is now costing you citations.

If the model cites Reddit for your topic, that’s an open door

This is the single most actionable diagnostic in AEO.

Ask an AI a core question in your category. If it answers by quoting a Reddit thread, a Quora post, or a random forum, it is telling you something specific: no credible brand has published a good answer to this. The model still has to answer, so it reaches for a conversation between strangers.

That is a content gap with a flashing light on it. Every Reddit citation in your category is a page you could own within a few weeks. Run your prompt index through two or three models, log every answer that leans on user-generated content, and prioritize those first.

The four layers that decide whether you get cited

When AEO fails, it usually fails at one of four layers — and teams tend to obsess over the last one while the first is broken.

1. Technical access. If AI crawlers can’t reach or parse your pages, nothing else matters. Check your robots.txt for blanket bot blocks, make sure key content isn’t locked behind client-side JavaScript, and confirm your pages render fast and clean without interaction.
2. Content structure. Models extract passages, not pages. Direct answers near the top, clear headings phrased as real questions, short self-contained paragraphs, and tables for comparisons all make your content easier to lift and cite.
3. Freshness and accuracy. Update pages when the underlying facts change — not on a calendar. Do not touch the publish date just to look current; models weigh substance, and stale-but-correct beats freshly-dated and unchanged. If a page is still earning citations, leave it alone.
4. Authority and corroboration. Models cross-check claims. Being right in one place is weaker than being right in several — your site, a YouTube explanation, a comparison roundup, an industry publication. Consistency across sources is what converts a mention into a recommendation.

Worth noting: traditional backlinks matter far less for AI citations than they do for classic rankings. They still carry weight in organic search, but if your AEO plan is a link-building plan, you are solving the wrong problem. For a broader view of the citation mechanics, see what it takes to get cited, and stay cited, in AI search.

Where to start if you’re a small team

Do not try to monitor everything. Teams that track thousands of prompts across five models end up with dashboards nobody reads and no decisions made.

Pick 20 to 30 prompts that map directly to revenue. Check them monthly, by hand if necessary. Fix the technical layer once. Then work the Reddit-citation gaps one page at a time. That is a realistic quarter of work, and it beats a comprehensive strategy that never ships.

Frequently asked questions

Is AEO different from SEO?

They overlap heavily but optimize for different outcomes. SEO targets a ranked position on a results page. AEO targets inclusion and framing inside a generated answer. Good technical SEO is a prerequisite for AEO, not a substitute for it.

How often should I check my AI visibility?

Monthly is enough for most businesses. Model outputs vary between runs, so daily checks mostly measure noise. Look for sustained directional change across a fixed prompt set.

Do backlinks help with AI citations?

Far less than with traditional rankings. Corroboration across independent sources matters more than link equity. Keep link building for organic search; don’t expect it to drive AI visibility.

Should I update publish dates to appear fresher?

No. Update the content when facts change and let the date follow. Date manipulation without substantive change is a pattern search systems already discount.

The short version

AI crawlers are taking more and sending back less, and that trade is not going to reverse. The brands that stay visible are the ones that stop optimizing for a click that no longer happens and start optimizing to be the source the answer is built from — accessible to crawlers, structured for extraction, corroborated across the places the model already trusts.