A client rang me in November last year, furious. His site was ranking #2 on Google for a fairly competitive SaaS term. Traffic was fine. But his main competitor, a scrappier site with half the domain authority, kept getting mentioned by name whenever anyone asked ChatGPT or Perplexity about the category. Not just cited. Named. Recommended. I spent two weeks pulling that competitor's content apart and what I found completely changed how I think about optimisation.
AI-native search is not a ranking system. It's a retrieval and synthesis system. And those two things need very different approaches.
Here's what I've pieced together after working through this on probably 40-odd client sites over the past year.
---
Why AI Search Works Differently to Google
Google ranks pages. Perplexity and ChatGPT Search source answers. The distinction sounds academic but it changes everything about what you should be doing.
When someone types a question into Perplexity, the system is looking for content it can quote with confidence. It wants specificity, clear attribution, and prose that holds up when extracted from its original context. A paragraph that reads well in isolation. A statistic with a named source. A definition that doesn't require the five paragraphs before it to make sense.
Google rewards authority signals built up over years. AI search rewards immediate legibility. The two overlap, but not as much as people assume.
Back in early 2024, Seahawk had a fintech client whose site was technically excellent. Fast, well-linked, strong E-E-A-T signals. Perplexity basically never cited it. We looked at the actual prose and found the problem in about ten minutes: every answer on the site was buried inside context-heavy editorial that you had to read from the top to understand. Standalone paragraphs meant nothing. We restructured three pillar pages over a fortnight and citations started appearing within six weeks.
---
Get Indexed by the Right Crawlers First
Before anything else, check whether Bing is crawling your site properly. This sounds odd if you've been living in a Google-first world, but ChatGPT Search is built on Bing's index. If Bing can't see you, ChatGPT almost certainly won't cite you.
Go into Bing Webmaster Tools and verify your site if you haven't already. Submit your sitemap. Check the crawl diagnostics. I've found maybe 30% of the sites we audit have either never submitted to Bing or have old sitemaps pointing to URLs that now redirect.
Perplexity runs its own crawler, PerplexityBot. Check your robots.txt file and make sure you're not accidentally blocking it. I've seen this exact mistake three times this year, once on a site that had been wondering for months why Perplexity never cited them despite ranking well organically.
The robots.txt check that takes 90 seconds
- Go to
yourdomain.com/robots.txtin your browser - Look for any
Disallowrules that apply toUser-agent: *or specifically toPerplexityBotorGPTBot - If you see blanket disallow rules from an old security plugin or migration, fix them
OpenAI publishes documentation on GPTBot if you want to see exactly how to allow or block it selectively.
---
Write Answers, Not Articles
This is the single biggest change I made to my own content workflow after that November client call. And honestly it took me longer than it should have to internalise it.
Most web content is written as a journey. Introduction, context, build-up, answer, conclusion. That structure works fine for someone who found your article via Google and is reading it top-to-bottom. It does not work for an AI that's scanning for a quotable response to a specific question.
AI models want what I call "answer-first paragraphs." The direct response to the implicit question comes in sentence one or two. The supporting context follows. Think of it like an inverted pyramid per paragraph, not just per article.
Compare these two openings for a paragraph about bounce rate:
Version A: "When users land on your website, their behaviour in the first few seconds tells you a great deal about whether your content is resonating. One of the most discussed metrics for understanding this is bounce rate, which..."
Version B: "Bounce rate is the percentage of sessions where a user views one page and leaves without interacting further. A rate above 70% on a content page usually signals a relevance mismatch between the ad or search result that brought them there and what they actually found."
Version B can be lifted and used as a citation. Version A cannot. Write Version B.
---
Structured Data Still Matters (Maybe More Than Ever)
I know structured data has been "important" for years and a lot of people have tuned it out. Fair. But for AI citation specifically, it's worth revisiting.
Schema.org markup gives AI systems a machine-readable summary of what your page claims, who wrote it, when it was published, and what organisation stands behind it. Perplexity in particular seems to weight author and organisation schema fairly heavily in my testing. A Person schema with a real jobTitle , a linked sameAs pointing to a LinkedIn or Wikipedia profile, and an Organization schema with a url and foundingDate isn't magic but it does signal that a real entity with accountable identity produced the content.
For Seahawk client work, I use Google's Rich Results Test to validate schema before deploying. Takes two minutes. Worth it.
The types that matter most for citation visibility:
ArticleorBlogPostingwithdatePublishedanddateModifiedPersonwithname,jobTitle,url,sameAsOrganizationwithname,url,logoFAQPagefor any page with a genuine Q&A structureHowTofor step-by-step instructional content
Don't fake it. If you publish an FAQ schema but the actual page doesn't have readable Q&A content, AI systems will figure that out quickly enough.
---
Build Topical Depth, Not Topical Width
Here's the thing about AI citation pools: they're small. Perplexity isn't pulling from the entire index equally. It pulls heavily from sources it's decided are authoritative on specific topics. And that authority isn't just domain-level. It's topic-cluster-level.
I've watched smaller sites get cited repeatedly on narrow topics simply because they had 15 well-written, tightly-focused pieces on that one subject. Meanwhile a larger site with 800 articles and maybe 5 on that topic got ignored entirely.
Pick the topic you actually want to own. Then build depth in that direction rather than breadth across ten different areas. For a B2B client last spring, we identified that their real differentiator was WordPress performance optimisation for WooCommerce stores. We stopped publishing generic "top 10 plugins" content and instead built out 18 focused pieces on that one narrow subject over three months. Perplexity citation appearances on related queries went from near zero to showing up in maybe 40% of test queries by month four.
What "topical depth" actually looks like in practice
It's not just publishing more articles. It's:
- Answering every obvious follow-up question within the cluster
- Linking between pieces consistently with descriptive anchor text
- Updating your cornerstone pieces every 90 days with fresh data or changed recommendations
- Including specific version numbers, tool names, and real-world outcomes rather than staying vague
---
Your Author Identity Needs to Exist Outside Your Own Site
AI systems don't just assess your site in isolation. They cross-reference. If I say I'm a London-based developer with 9 years of experience and they can find corroborating signals on LinkedIn, on podcast appearances, on guest posts, in press mentions, the claim becomes credible. If the only place that claim exists is the About page on my own site, it's much weaker.
This is why I've been quietly pushing clients to do things that feel more like PR than SEO. Guest posts on industry sites. Getting quoted in round-up articles. Podcast appearances, even small ones. Wikipedia edits in your area of expertise (genuinely, where appropriate). LinkedIn content that gets indexed.
None of this is new. But it matters more now because AI systems are doing a version of entity resolution: they're trying to confirm that the person or organisation making a claim is real and credible, not just that a page about that person or organisation exists.
For what it's worth, I started writing more on my personal site in late 2023 specifically for this reason. Not for traffic. To build a trail of indexed, attributable opinions that AI systems could find when checking whether "Gautam Khorana from Seahawk Media" is a real person with real expertise.
---
Freshness Signals and the Update Cycle
Perplexity in particular tends to favour recently updated content on fast-moving topics. I've noticed this pattern clearly with anything touching AI tools, SaaS pricing, or regulatory changes. An article from 2022 that hasn't been touched since, even a very good one, loses citations to a mediocre one that was updated three weeks ago.
Set a calendar reminder. Every 90 days, go back to your top 10 pages and update something real: a statistic, a tool recommendation, a pricing figure, a date reference. Change the dateModified in your schema too. Don't just make cosmetic edits. Actually improve the content.
If you're using WordPress, the PublishPress Revisions plugin is decent for managing this. Or just build it into a content audit spreadsheet. Low-tech works fine.
---
FAQ
Does domain authority still matter for AI citations?
It's a factor but not the dominant one. I've seen DA 20 sites get cited over DA 70 sites on specific queries because the smaller site had clearer, more quotable answers on that exact topic. Established authority helps get you in the consideration pool but it won't save vague or poorly structured content.
Should I be writing specifically for ChatGPT versus Perplexity, or is one strategy enough?
One strategy covers most of it. The principles are the same: clear answers, real author identity, proper crawl access, topical depth. The main practical difference is that ChatGPT leans on Bing's index more heavily, so Bing Webmaster Tools matters more there. For Perplexity, the freshness signal seems stronger. Otherwise, optimise once and it carries across both.
How long does it take to see results?
Honestly, varies wildly. The fintech client I mentioned took six weeks after restructuring. The WooCommerce client took four months of consistent publishing. I've also seen sites start getting cited within three weeks of fixing a blocked crawler. There's no reliable timeline. Do the work, check your progress with manual test queries every few weeks, adjust.
What tools can I use to check if I'm being cited?
Manual querying is still the most reliable method. Use a consistent set of 10-15 test questions in your topic area and run them in Perplexity and ChatGPT Search weekly. For scale, some agencies are starting to use tools like Brandwatch or even custom scripts that hit the Perplexity API to track citation frequency. Nothing fully automated exists yet that I'd call production-ready.
My site is being cited but not for anything useful. How do I fix that?
Look at which pages are getting cited and what question the AI is answering when it pulls them. Usually the page is ranking for the wrong query because it's topically vague. Tighten the focus. One clear topic per page, one clear answer per section. The more surgically specific a page is, the more likely it gets cited for the right thing.
---
Nine years in, 12,000-odd sites later, and the fundamentals are still fundamentals. Be legible. Be credible. Be findable. The medium changes. The underlying logic doesn't.