Citation engineering: how to get cited by ChatGPT, Perplexity, and AI Overviews

Users who arrive on your site through a ChatGPT or Perplexity citation convert at 3-4x the rate of users who arrive via organic search. The model did the vetting work. The user lands pre-sold.

Yet most marketing teams cannot tell you what their citation share is on those engines this week. They can read off Google rankings to two decimal places. The new traffic source, the one with multiplicatively better conversion, is unmonitored, unowned, and quietly decaying.

Closing that gap is what citation engineering is. It is the operating discipline of getting your brand reliably cited across the answer engines, on the surfaces they pull from. It overlaps with SEO. It is not SEO.

AEO is not SEO with a new acronym

SEO ranks a page so a human clicks. Answer Engine Optimization gets your brand quoted inside an AI-generated answer the human reads instead of clicking. The disciplines share inputs. The execution does not rhyme.

Three things have to land for an answer engine to cite you:

• Retrieval. Your content has to surface in the engine's grounding step. That is crawl access, structured data, freshness signal, and inbound links.
• Selection. The model has to pick your sentence over a competitor's. That is how directly you answer the literal question, how trustworthy the source looks, and when the page was last touched.
• Attribution. The citation has to appear as a clickable link to your domain. That layer varies by engine and is the part most teams never measure.

Each layer has different failure modes. Pages not refreshed at least quarterly lose AI citations roughly three times faster than pages that are. Most teams optimize the retrieval layer, only for Google, on a once-a-year refresh cadence.

Meanwhile, roughly 60% of US and EU searches now end without a click thanks to AI Overviews. The traffic is not gone. It moved into a layer your team is not measuring.

The citation surface is wider than your blog

Your owned pages are one of five surfaces answer engines pull from. The other four are where most teams quietly hemorrhage share. The mix is also wildly different across engines, so a single citation strategy will not cover all three at once:

• Owned pages. Your blog, docs, and category pages. The only surface most teams actively work.
• Reddit and community threads. Perplexity is famously Reddit-heavy, with Reddit reportedly accounting for nearly half of its top citations, and Google AI Overviews lean on Reddit for product, comparison, and 'best of' queries. A single thread can shift a brand's citation share more than a quarter of blog work.
• Listicles and round-ups on third-party publications. Industry analyses estimate roughly 85% of AI brand mentions originate from third-party sources, not your domain. If you are not in the comparison posts, you are not in the answer.
• Encyclopedic sources. ChatGPT pulls disproportionately from Wikipedia and reference content, with some studies pegging it at nearly half of top citations. If your brand has no Wikidata entry or a thin one, the model has to guess at your identity.
• Reviews and Q&A communities. Quora, Stack Exchange, and niche forums. The long-tail intent layer where engines find the actual phrasing of user questions.

A citation engine works all five surfaces. A blog calendar works one.

Why this breaks traditional SEO teams

Citation engineering is a systems job, not a content job. To do it well, you need a single operator who can:

• Pull current citation share from each engine, weekly, on a defined query set.
• Cross-reference that share with Search Console clicks to see which queries are decoupling from Google ranking and which are not.
• Identify which third-party listicles, Reddit threads, and YouTube videos drive the bulk of existing AI mentions.
• Refresh on-page assets that are losing citations to freshness decay.
• File mention requests, pitch listicles, and seed authentic Reddit answers where the brand is missing.
• Patch schema, fix entity graphs, and keep Wikidata in sync.
• Do all of that every week, across every product category, without dropping any of it.

That is six tools, five surfaces, a writing job, and a relationships job. No in-house marketing team I have met runs this end to end. The teams that look like they are running it are quietly running an agency under the hood.

How Oz runs a citation engine

Oz is Claude Code for marketers. The same shape AI coding tools brought to engineering, an agent with access to your full toolchain, making decisions, shipping work end to end, applied to marketing surfaces.

For citation engineering specifically, Oz agents wire into your Google Search Console, Google Analytics, and DataForSEO accounts plus a Reddit and a citation-monitoring layer, and operate across the action categories the team composes:

• Citations. The agent watches share-of-citations across ChatGPT, Perplexity, and AI Overviews on a defined query set, surfaces the queries where you are losing ground, and writes the on-page or off-page action that closes the gap.
• Reddit engagement. It scans threads in your category for missing or wrong-by-omission mentions, drafts authentic, value-first answers consistent with your brand voice, and queues them for human review before posting.
• Off-page. It pitches listicles, files mention requests, and tracks third-party coverage as a measurable surface, not a vanity metric.
• On-page. It refreshes pages losing citations to staleness, tightens the answer paragraphs engines actually quote, and patches schema and entity markup.

The team's job stops being 'ship a blog post.' It becomes 'watch citation share and approve the agent's actions.' That shift is the same one agentic marketing in general is forcing: the executor role compresses, the orchestrator role expands.

What it isn't

Citation engineering is not a scoring problem. It is an action problem distributed across surfaces no single tool owns.

It is not Surfer or Clearscope. Those tools score on-page content against a target keyword. They do not refresh, they do not pitch, they do not post on Reddit, and they do not monitor citations.

It is not Jasper or Copy.ai. Those write copy. Citation share is not a copy problem. The teams losing it are losing it on freshness, structured data, and third-party presence, not on prose quality.

It is not a horizontal agent builder like AirOps or Relevance AI. Those give you a toolkit to wire any agent. Citation engineering is opinionated. The agents have to know what GSC means, what a structured FAQ schema is, why a Reddit moderator will remove a comment that smells like a brand, and how to pace mention pitches so they do not all land in the same news cycle.

What this means for your team

Citation share is going to become a top-line marketing metric inside 18 months. The teams that can answer 'what is our share of citations on these 200 high-intent queries, and what changed this week' will pull away from the teams still measuring rankings only.

We have already watched this curve, in the agency work behind Oz, on the brands where we got there first. The same playbook that compounded $1B+ in incremental organic revenue across 500+ brand portfolios at Passionfruit, with HP, Unilever, L'Agence, Solawave, Goli, and Mokobara among them, is the spine of how Oz runs citations now. The lift is not a content lift. It is a systems lift. You either staff a five-person ops team to run it, or you give one operator an agent.

If you want to be the operator running the agent, join the Oz waitlist. The first cohort gets the citations agent on day one.