How to get your agency cited by ChatGPT (for free)

Why your agency's own website almost never gets cited for the queries that matter

Short answer

Yes, ChatGPT and Perplexity can and do cite official company websites, but it depends heavily on what is being asked. For questions about your pricing, services, or documentation, your own site is often the preferred source. For comparison and recommendation questions, the ones that actually bring in new business, third-party sources take over almost entirely.

The longer answer

No outside party can see exactly how these platforms weight every source internally. What is verifiable is what they cite, and that pattern is consistent. A 2026 study tracking 1,016 citation records across Perplexity and Google AI Overviews, built around recommendation-style queries, found official company pages accounted for just 1% of Perplexity citations and 4% of Google AIO citations.^[1] ChatGPT, in the same study, provided no source URLs in any of its 40 tested responses. A separate Ahrefs analysis of 1.4 million ChatGPT citations found a related pattern: ChatGPT typically retrieves dozens of URLs per query but ends up citing roughly half of them, and official pages rarely survive that cut when the query is comparative rather than factual.^[2]

Where official sites do well, and where they do not

• For pricing, documentation, and direct factual questions, your own site is a strong source.
• For comparison questions ("who is the best agency for X"), third-party sources dominate.
• For reviews and reputation questions, independent platforms are preferred over your own claims.
• For regulatory or financial questions, official bodies take precedence over either.

What this means for your strategy

If a buyer already knows your name, your own site likely gets cited fine. If a buyer is still deciding who to choose, which is the moment that actually drives new enquiries, your own site is rarely where these platforms look. That is the gap third-party coverage closes.

How does ChatGPT actually decide what to mention?

Short answer

ChatGPT generally cites a source when it has searched the live web and a specific page directly supports a claim in its answer. It does not cite when it is answering from general training knowledge, common facts, or its own reasoning.

What drives the decision

• Relevance. The source has to answer the exact sub-question being asked, not just the general topic.
• Authority. Stronger domains and pages backed by original data or clear factual support are favoured over promotional or vague pages.
• Freshness. For anything time-sensitive, recently updated pages are more likely to be pulled in.
• Structure and extractability. Pages with clear headings, concise definitions, and statements that can be lifted as a standalone fact are easier for the system to use. A broad about page rarely qualifies. A page that states a specific fact plainly usually does.

Why a page can be searched and still skipped

ChatGPT typically retrieves far more pages than it ends up citing. A page gets dropped when it is redundant with a stronger source, too generic to support a specific claim, or simply not the best match for the exact thing being asked. When several sources say the same thing, the system tends to keep only the clearest one. That is consistent with the separate finding that the large majority of ChatGPT citations come from pages well outside the top 20 Google results. Ranking position is not the gate. Having the clearest extractable answer is.

What this means in practice

Two things have to be true at once. The source has to be one ChatGPT trusts enough to retrieve in the first place, which comes back to the third-party coverage point above. And the page itself has to state something specific and liftable rather than burying the answer in general explanation. Getting found and being citable are two separate problems, and most agency content only solves the second one, if that.

Why do ChatGPT, Perplexity, and Google AI Overviews cite different sources?

Short answer

These are not variations of the same system. An analysis of 680 million AI citations found only 11% domain overlap between ChatGPT and Perplexity, a figure independently confirmed at 12% by a separate cross-engine study.^[3] A strategy built around one platform leaves the other 89% of its citation landscape unaddressed.

How each platform sources differently

ChatGPT draws primarily from reference-style sources accumulated during training, with live web retrieval only when the query triggers it. Inline citations are inconsistent outside Deep Research mode, and with fewer than 7 sources per answer on average, each cited position is genuinely competitive.

Perplexity searches the live web for almost every factual query and cites sources inline by default. Its preference for recently updated, community-sourced content means citation share fragments faster, so consistent presence across multiple placements matters more than a single strong one.

Google AI Overviews has the strongest correlation with organic ranking of the three, though citations still frequently come from pages well outside the top ten results. Video and forum content appear disproportionately often relative to their organic ranking positions.

Where this gets complicated

The platforms not only prefer different sources, they can reach similar-sounding conclusions from almost entirely separate evidence pools. Two agencies can both appear to "do well in AI" while being cited by completely different platforms from completely different source types. That makes single-number AI visibility reporting genuinely misleading rather than just incomplete.

What this means in practice

Getting cited by ChatGPT and getting cited by Perplexity require different approaches. Wikipedia-adjacent authority and established reference coverage moves the needle on ChatGPT. Fresh, specific, community-visible content moves it on Perplexity. For most agencies, one of these is significantly easier given where they already have coverage, which is worth knowing before deciding where to put effort first.

Does schema markup actually drive AI citations?

Short answer

Schema markup does not directly cause ChatGPT or Perplexity to cite your page. The evidence on this is consistent across multiple large-scale studies. What drives citation is third-party coverage on sources these platforms already trust, not structured data on your own site.

What schema actually does and does not do

• Comprehension, not citation. Schema helps AI systems identify what a page is about and how content pieces relate. It is a comprehension signal, not a citation trigger. An Ahrefs study tracking 1,885 pages that added JSON-LD schema found it did not boost citations on ChatGPT or Google AI Mode.^[4]
• Generic types do nothing measurable. Organisation and Article schema, the types most commonly recommended in GEO guides, showed no meaningful predictive power in citation studies. More attribute-rich types like Product or Review performed marginally better, but still well below content and authority signals.
• The correlation trap. Sites with schema are often cited more, which is where the advice comes from. The reason is that those sites also tend to have stronger content and higher domain authority. The AI cites them for those reasons, not for the structured data itself.

Why this matters more than it sounds

We saw this directly in the 2026 AI Citation Visibility Study across 40 DeFi protocols. Pendle was the only protocol in the dataset with full JSON-LD schema implementation. It scored zero citations across all tested queries. Aave, with no schema at all, ranked second. If schema were a meaningful citation driver, that result would not be possible. It confirms what the Ahrefs data shows at scale: schema correlates with citation because it correlates with better sites, not because it causes citation on its own.^[1]

What actually drives citation visibility

Third-party coverage is the primary lever. For ChatGPT that means Wikipedia-adjacent authority, established reference sources, and well-indexed directories. For Perplexity it means recently updated content on community platforms and specialist publications. For Google AI Overviews it means pages already performing in organic search, particularly those with clear direct answers near the top.

How to structure your content so AI can quote it

Short answer

Put a direct 2-3 sentence answer at the top of every page you control, written so it can be lifted whole. This is the single most consistently recommended on-page change across every platform that retrieves and cites content, and unlike schema, it directly matches how AI systems extract passages rather than how search engines rank pages.

What makes content extractable

• Direct answer first. The main answer goes in the first two to three lines, before any context or framing. AI retrieval systems pull from the top of pages disproportionately. Content that buries its answer in the third paragraph rarely gets cited even when the answer itself is good.
• Question-matching headings. Headings written as natural questions improve how AI systems chunk and retrieve sections. The heading becomes the query, the paragraph underneath becomes the answer.
• Short, scannable paragraphs. Bullets, numbered steps, tables, and paragraphs under 50 words are easier to extract than dense prose. Each paragraph should carry one point, not three.
• Specific facts over general claims. "Many agencies struggle with AI visibility" is not extractable. "A 2026 study tracking 1,016 citation records found official company pages accounted for just 1% of Perplexity citations" is. Named entities, dated statistics, and concrete definitions give AI systems something to lift and attribute.
• Consistent terminology. Using the same phrase for the same concept throughout a page helps citation systems match that concept reliably across queries. Switching between different terms for the same idea fragments the signal.

Where this gets misapplied

The answer capsule is an on-page technique. It improves how a page performs once it is already in the pool of sources a platform considers. It does not get you into that pool in the first place. Most agency websites applying these techniques are solving the second problem while the first one remains untouched, which is why on-page GEO work often produces no measurable citation change. The source eligibility problem has to be addressed before on-page extractability becomes relevant.

What this means in practice

Check every service page and key landing page for two things: does it lead with a direct answer, and is the answer specific enough to be attributed rather than generic enough to be ignored. For most agency sites the answer to both is no, which is fixable in an afternoon without touching the site's architecture or authority profile.

Does fresher content get cited more often?

Short answer

Citation visibility drifts. Pages that appear consistently in AI-generated answers can drop out of rotation as fresher, more specific content appears elsewhere. Doing the work once and leaving it is a reasonable starting point, but not a maintenance strategy.

What the data says about freshness

• The 393-day preference. An Ahrefs study found ChatGPT shows the strongest preference for content around 393 days old, roughly 13 months, not yesterday. The pressure to publish constantly to stay cited is overstated.^[5]
• The 50% threshold. A separate analysis found 50% of AI-cited content was updated within the last 13 weeks. Pages not refreshed quarterly were three times more likely to lose citations than those that were.
• The 67% freshness lift. Content updated within three months averaged 6 AI citations versus 3.6 for outdated pages in one cross-platform analysis, a 67% gap attributed to freshness alone.

Where the panic narrative goes wrong

The most-shared claim on this topic is that 74% of ChatGPT citation sources change weekly, a figure that originated from a single YouTube video with no published methodology. It circulates because it is alarming, not because it is verified. The same source, in a separate post, noted that the average ChatGPT-cited page is 500 days old. Both cannot be true simultaneously. The weight of more rigorous evidence points toward a recency preference measured in months, not days, with quarterly refreshes as the practical minimum rather than constant republishing.

What this means in practice

Treat pages that matter for citation visibility as living documents, not published artefacts. A page covering your agency's key services, updated once a quarter with a current statistic or a corrected claim, will hold its position more reliably than one left untouched since publication. That is a small, recurring maintenance task, not a content production programme.

What does the evidence say about AI visibility across industries?

Running AI visibility research across two completely different industries, DeFi protocols and UK recruitment agencies, produced a finding that was not expected. The results are nearly identical.

The source types dominating citation for a blockchain lending protocol are the same source types dominating citation for a specialist engineering recruiter. Third-party aggregators, community platforms, industry directories, and independently published content account for the overwhelming majority of citations in both cases. Official websites, whether a protocol's documentation or an agency's service pages, appear in a small fraction of results regardless of how well-built or well-optimised those pages are.

The problems agencies raise in early conversations are the same problems crypto protocol teams raise. Why is our website not cited when someone asks about us? Why is a competitor mentioned and we are not, even though we are larger? Why does our content rank in Google but not appear in AI answers? The fixes are the same too: third-party coverage on the source types each platform trusts, content structured for extraction rather than reading, and a clear picture of which queries are actually driving citation in the category before doing anything else.

That consistency across two unrelated industries suggests this is not a niche problem specific to crypto or to recruitment. It is a structural feature of how AI platforms build answers. They synthesise from the sources they already trust, and for most professional service categories, those sources are not the company's own website.

The free path described in this guide, third-party coverage on the right sources, content structured for extraction, and a quarterly refresh cycle, will move the needle for most agencies without spending anything. Where it gets harder is knowing which specific queries are driving citation in your category, which sources are already being cited for those queries, and exactly where your agency sits relative to competitors across each platform. If you would like to talk through where your agency currently stands, get in touch via the contact page. At minimum there are usually a handful of simple, free adjustments specific to your situation that are worth knowing about. For agencies with a more significant visibility gap, a manually researched visibility review can identify exactly what is missing and the most direct path to closing it.