In this article
A prospect types a question into ChatGPT instead of Google: “best accounting software for a 10-person agency” or “who handles DMCA takedowns for small publishers.” The model answers with three or four named companies and a short reason for each. Your company ranks on page one of Google for the same query. It isn’t in the answer.
That gap is what answer engine optimization (AEO) and generative engine optimization (GEO) are built to close. Neither is a replacement for SEO. Both depend on a different set of mechanics, and getting cited requires understanding those mechanics rather than repackaging a blog strategy with a new label.
Why this is a different problem than ranking on Google
Google’s results page still works the way it has for two decades: ten blue links, ranked by a mix of relevance and authority signals, with the user doing the reading. ChatGPT and Perplexity compress that step. They read for the user and hand back a short, synthesized answer with, sometimes, a citation attached.
That synthesis step changes what matters:
- The unit of value isn’t a ranking position, it’s a sentence. A model pulls the one paragraph on your page that answers the question cleanly, not your whole page.
- Citations are selective, not exhaustive. Perplexity typically shows three to eight sources per answer. ChatGPT’s browsing and search modes show fewer. If your page says the same thing as five competitors but says it less directly, it gets skipped.
- Freshness and retrieval replace some of what links used to do. Perplexity runs its own crawler and index. ChatGPT’s search features draw on Bing’s index and real-time retrieval. Neither relies purely on backlink authority the way classic Google ranking does.
None of this means links and domain authority stop mattering. Both tools still favor pages that already rank well and get referenced elsewhere. A second layer sits on top: can the system actually extract a clean, quotable fact from your page once it finds it?
The crawlers you’re optimizing for
A handful of named bots decide whether your content even gets read:
- GPTBot and OAI-SearchBot, OpenAI’s crawlers, documented in OpenAI’s own crawler disclosures.
- PerplexityBot, Perplexity’s crawler, which the company has publicly described in its own documentation.
- Google-Extended, a separate signal from standard Googlebot that controls whether Google’s AI features can use your content.
- Bingbot and CCBot (Common Crawl), indirect paths, since several AI systems train on or retrieve from these indexes.
Checking your server logs or a tool like Google Search Console’s crawl stats for these user agents tells you something SEO dashboards won’t: whether the AI tools are even showing up. If GPTBot and PerplexityBot never hit your site, nothing downstream matters yet. If they’re crawling and you’re still not cited, the problem is further down the pipeline.
Make the page readable by a machine first
Most sites write for a person scanning top to bottom, with the key fact buried under three paragraphs of scene-setting. A retrieval model doesn’t scan. It needs the answer close to the question.
Three changes do most of the work:
Answer the question in the first sentence of the relevant section. If a page is titled “Pricing,” the first sentence under that heading should state a number or a range, not a sentence about your philosophy on fair pricing.
Use schema.org structured data where it genuinely applies. FAQPage, Product, Organization, and Review markup give crawlers an explicit, machine-parseable version of what’s already on the page. This doesn’t trick anything into citing you, it removes ambiguity about what the page is claiming.
Keep a clean, current sitemap and don’t block useful crawlers in robots.txt. Some sites, trying to block AI training, accidentally block GPTBot or PerplexityBot entirely with a blanket disallow rule, which also removes them from citation in search-style answers. Decide deliberately whether you want to allow retrieval but block training (OpenAI and others support this distinction in their documentation) rather than blocking everything by default.
A newer, still-informal convention worth knowing about is the llms.txt file, a plain-text summary of a site’s key pages, proposed as a lightweight way to help AI systems find the right content fast. Adoption is uneven across the web right now, so treat it as a nice-to-have, not a guaranteed lever.
Write the sentence that gets lifted
Technical plumbing gets you read. Content gets you quoted. The sentences that end up in an AI answer tend to share a shape:
- A specific number with context: “Setup takes 10 business days and includes migration of up to 5,000 records,” not “fast and easy setup.”
- A named entity instead of a vague category: “integrates with QuickBooks and NetSuite,” not “integrates with popular accounting tools.”
- A direct comparison, stated plainly: “Plan A includes phone support; Plan B does not.”
This is also where FAQ pages earn their keep, beyond the schema markup. A well-formed FAQ, real questions a buyer actually asks, answered in two or three sentences each, is close to the exact format a model needs to lift and attribute. Pad it with marketing language and the model has to do extra work to find the fact, which pushes it toward a competitor’s cleaner page instead.
One caveat applies directly to this kind of content: if you publish comparison pages, “best of” lists, or customer testimonials to get cited, the Federal Trade Commission’s endorsement guidelines require clear disclosure of any paid or incentivized relationship behind them. A comparison page secretly written by one of the vendors it ranks first is a disclosure problem before it’s an AEO problem.
Where citations actually come from
Neither ChatGPT nor Perplexity cites company websites exclusively. Third-party sources carry real weight:
| Source type | Why it gets cited | Example platforms |
|---|---|---|
Review and comparison sites |
Structured, comparative, frequently updated |
G2, Capterra, TrustRadius |
Community discussion |
Seen as unfiltered, specific opinion |
Reddit, niche forums |
Reference and encyclopedic sources |
High trust weighting in most retrieval systems |
Wikipedia |
Press and trade publications |
Timestamped, editorially vetted |
Industry trade press |
Your own site |
Primary source, but only if the fact is extractable |
Pricing pages, FAQ pages, documentation |
A company with a thin review footprint on G2 or no presence in the forums where its buyers actually ask questions is handing those citation slots to competitors by default, no matter how clean its own website is.
Checking whether it’s working
There’s no equivalent yet of Google Search Console for AI citations, though several third-party trackers have started filling the gap. In the absence of a single dashboard, three manual checks get you most of the way:
- Run your 10 to 15 highest-intent buyer questions through ChatGPT and Perplexity monthly and record whether your company appears, with what phrasing, and from which source.
- Check server logs or a log-analysis tool for GPTBot, OAI-SearchBot, and PerplexityBot hits, and watch which pages they request most.
- Track referral traffic from chat.openai.com and perplexity.ai in your analytics, small volume right now for most companies, but a leading indicator of whether the citations are converting.
A note from our side on building this kind of infrastructure
We don’t sell AEO audits as a packaged product. What we do build, regularly, is the infrastructure this whole approach depends on: clean content management systems, structured data pipelines, and API integrations that make a site’s facts extractable in the first place.
One relevant example from our own operation: we run a cold outreach system where, before writing anything to a lead, a model reads that company’s homepage and up to four internal pages, about 15,000 characters, and extracts one or two concrete sentences about what that business actually does. It only works when the target page states something specific and plain. Vague copy produces a vague, useless output. That’s the same filter an AI crawler applies to your own site when deciding what’s worth quoting: specific beats polished every time.
We’ve also built systems that continuously monitor search results across multiple engines on a client’s behalf, a dashboard and bot that track a brand’s presence across Google, Bing, Yahoo, and Yandex and flag what shows up. The engines are different from the chat-based tools discussed here, but the underlying discipline, watching what a machine surfaces about you and reacting fast, carries over directly.
On the content-infrastructure side, we built a casting platform from scratch with its own admin panel and a dedicated blog section designed for ongoing SEO publishing, reaching a top search position for its own brand name within about a week of launch. The client was advised to keep publishing several articles a month to hold that position, the same cadence that keeps content fresh enough for retrieval-based tools to keep pulling from it.
The technical rebuild most sites need before any of this works, structured headings, FAQ schema, a clean sitemap, pages that state facts instead of hedging, fits inside a fixed-price build. At Pazl that starts around $3,260–$6,520 (≈€3,000–€6,000), scoped after we’ve reviewed the actual site, with a six-month guarantee on what gets delivered.
FAQ isn’t optional, it’s part of the build
Treating the FAQ section as an afterthought defeats the purpose of everything above it. It should be written with the same discipline as the rest of the page: real buyer questions, direct answers, FAQPage schema attached, and no marketing padding between the question and the answer.
Frequently asked questions
How long does it take before ChatGPT or Perplexity start citing us after making these changes? There’s no fixed timeline, since it depends on how often each tool’s crawler revisits your site and how competitive your topic is. Perplexity’s own crawler tends to pick up changes faster than retrieval tied to a less frequently refreshed index. Expect weeks rather than days for a first citation, and treat it as an ongoing practice rather than a one-time fix.
Do we need to rebuild the entire website, or can we start with a handful of pages? Start with the pages that answer your highest-value buyer questions, usually pricing, comparison, and FAQ pages, rather than the whole site at once. Those are also the pages most likely to already have scattered, usable content that just needs restructuring instead of a full rewrite.
Will optimizing for AI answer engines hurt our regular Google rankings? No. The technical changes involved, clear structure, schema markup, direct answers, a clean sitemap, are the same fundamentals that help traditional SEO. Nothing in this approach asks you to write worse copy for human readers to please a machine.
What do you need from us to scope a project like this? Access to your current site (or at minimum a list of your highest-traffic and highest-intent pages), a sense of the 10 to 15 questions your buyers actually ask before purchasing, and whichever CMS or hosting setup you’re currently running. We work from there to decide what’s a content fix versus a structural rebuild.
Can you verify whether our site is already being crawled by these AI tools? Yes, through server log analysis looking for GPTBot, OAI-SearchBot, and PerplexityBot requests. If none show up, that’s the first problem to solve, before any content rewrite.
If any of this sounds like the gap between what your site says and what an AI tool can actually pull from it, write to us at hello@pazl.ai and we’ll take a look at where the disconnect is.



