All Articles

How does AI decide what to cite, and how do you get content cited?

A
AISO Studio
||9 min read

How ChatGPT chooses sources involves evaluating content for structure, authority signals, named citations, and direct-answer formatting. The engines have not published their full selection criteria, but patterns emerge when you test the same query across ChatGPT, Perplexity, and Google AI Overviews. This guide gives you the checklist and testing method to make client content citable.

What You'll Be Able to Do

After reading this, you will have a complete citability scoring rubric you can run on any client page. You will know which structural elements correlate with being cited. You will have a testing prompt to check whether a client's site appears in AI responses. And you will know where citability breaks down so you can fix it before the client asks why their competitor got cited instead.

Prerequisites

You need access to the client's published content. You need a way to check whether that content includes named sources. You need to be able to view the page's HTML or use a browser inspector to check for schema markup. If you are testing citations, you need accounts on ChatGPT, Perplexity, or access to Google Search with AI Overviews enabled.

How AI Engines Select Sources

AI engines extract content from pages they can parse, verify, and attribute. They prefer pages that name their own sources, use structured data, and answer questions directly in the first paragraph. They avoid pages with unsupported claims, vague attribution, or content that requires interpretation.

The engines do not publish their ranking algorithms. What follows is based on testing client pages and noting which structural elements appear consistently in cited content.

Why Structure Matters

AI engines parse content in blocks. A paragraph that answers a question in 40-60 words can be extracted and cited as-is. A 200-word paragraph with three ideas cannot. The engine has to decide which sentence answers the query, and if it cannot decide cleanly, it moves to the next source.

Short paragraphs with one idea each make extraction reliable. Headers that match common questions make the content discoverable. Bullet lists and numbered steps make the structure explicit.

Why Named Citations Matter

When your client's page cites a named source, the AI engine can verify the claim. When the page says "research shows" or "studies indicate" without naming the source, the engine has nothing to check. Content with named citations gets treated as more reliable.

This is the single biggest factor you control. Every claim that carries evidence should name who it came from. If the client cannot name a source, rewrite the sentence without the claim.

Why Freshness and Authority Signals Matter

AI engines prefer pages that show recent updates, clear authorship, and domain authority in the topic. A page with a publish date, an author byline, and inbound links from related sites signals that the content is maintained and trusted.

Freshness is defined as the last time the page was updated, not the original publish date. If the client published a guide some years ago and has updated it since, the updated date is what matters. Add a "Last updated" line at the top of the page.

Why Direct-Answer Formatting Matters

AI engines extract the first paragraph when it answers the query directly. If the first paragraph is a question, a story, or background context, the engine has to scan further down the page. That increases the chance it will cite a competitor's page instead.

The first paragraph should state the answer in 80-200 characters. Use a definition verb: "is," "means," "refers to," "involves." Then expand in the paragraphs that follow.

Common Mistakes

Agencies often write content that reads well but cannot be cited. Here are the patterns that break citability:

  • Long introductions before the answer. The client wants to set context, but the engine wants the answer in the first paragraph. Move context to a second section.
  • Unsupported claims. Sentences that state percentages, outcomes, or general truths with no named source. If your client's page includes a claim like "most businesses see a 30% increase" without attribution, the engine has no way to verify it, and an unverifiable claim gives it no reason to cite the page.
  • Vague attribution. Phrases like "According to industry data" or "research shows" with no organization named score zero for citability. Name the source or remove the claim.
  • No schema markup. The page has good content but no structured data to tell the engine what it is looking at. Add Article, FAQPage, or HowTo schema.
  • No author or date. The engine cannot assess freshness or authority. Add both.

Citability Checklist and Scoring Rubric

Use this rubric to score any client page for citability. Each factor is worth points. A page needs 70+ points to be consistently cited.

Structure (30 points)

  • First paragraph answers the query directly (10 points). 80-200 characters, uses a definition verb, ends with a period.
  • Paragraphs are 40-60 words (5 points). One idea per paragraph.
  • Headers match common questions (5 points). Use H2 for main questions, H3 for follow-ups.
  • Bullet lists or numbered steps present (5 points). Makes structure explicit.
  • FAQ section with Question: format (5 points). Uses "### Question:" for each Q&A pair.

Citations and Evidence (25 points)

  • At least 3 named sources (15 points). "According to [Organization]," or "[Organization] reports that."
  • No unsupported claims (10 points). Every claim that states a statistic, outcome, or general truth must name the source. If your client cannot verify the source, remove the claim rather than use phrases like "research shows," "studies indicate," or "most businesses."

Authority Signals (20 points)

  • Author byline present (5 points). Name and role or organization.
  • Publish or last-updated date visible (5 points). At the top of the page.
  • Inbound links from related sites (5 points). Check backlink profile.
  • Domain has topical authority (5 points). Other pages on the site cover related topics.

Schema Markup (15 points)

  • Article schema present (5 points). Includes headline, author, datePublished, dateModified.
  • FAQPage schema if FAQ section exists (5 points).
  • HowTo schema if steps are present (5 points).

Freshness (10 points)

  • Updated within the last 12 months (10 points). Or 5 points if updated within 24 months.

Total: 100 points

A page scoring 50-69 points will be cited occasionally. If your client's page scores below 50, expect citation frequency to drop.

Testing Prompt for Client Sites

Use this prompt in ChatGPT, Perplexity, or any AI engine that cites sources. Replace the bracketed fields with specifics.


Prompt:

"[Question the client's page answers]. Please cite your sources and include the full URL for each source you reference."

Example:

"What is the best way to structure a SaaS pricing page? Please cite your sources and include the full URL for each source you reference."


Run this prompt and check whether the client's page appears in the citations. If it does not, run the citability checklist above and note which factors are missing. Fix those factors and test again in 2-4 weeks.

Troubleshooting

The client's page scores 70+ but is not being cited

Check whether the page is indexed. Search "site:[client domain] [page title]" in Google. If it does not appear, the page is not indexed. Submit the URL in Google Search Console.

Check whether the page has a robots.txt block or noindex tag. View the page source and search for "noindex." If present, remove it.

Check whether the page is new. It can take 2-4 weeks for AI engines to index and cite new content. If the page is older than 4 weeks and still not cited, check the backlink profile. Pages with no inbound links are less likely to be cited.

The client's page is cited but the citation is incomplete or wrong

This happens when the first paragraph is vague or when the page has multiple competing answers. Rewrite the first paragraph to answer the query in one clear sentence. Remove any competing definitions or answers from the top of the page.

The client's page is cited for the wrong query

The page's header or first paragraph matches a different query better than the intended one. Rewrite the header and first paragraph to match the intended query exactly. Use the exact phrasing the client's audience types into the search bar.

The page has named citations but still scores low

Check whether the citations are verifiable. If the page cites "a recent study" or "industry research" without naming the organization, it does not count. Replace vague citations with named sources or remove the claim.

Frequently Asked Questions

Question: What are citations for SEO and how do they differ from AI citations?

Citations for SEO refers to mentions of a business name, address, and phone number on other sites, used primarily for local search. AI citations meaning is different: it refers to when an AI engine names and links to your content as a source in its response. AI citations require named sources and structured content. SEO citations require consistency across directories.

Question: How do I write content for AI search versus traditional search engines?

To write content for search engine ranking, you target keywords and build backlinks. To write content for AI search, you answer questions directly in the first paragraph, name your sources, and use schema markup. AI-friendly content is a form of structured, verifiable content that AI engines can extract and cite. Both approaches overlap, but AI search prioritizes structure and attribution over keyword density.

Question: Does AI content affect Google rankings?

No study has shown that AI-generated content is penalized by Google if it is factually accurate and useful. Google's published guidance says content quality matters, not how it was produced. The question is whether the content is verifiable, well-structured, and answers the query. If you are looking for an AI content Google rankings study, none has been published by Google or a named research organization as of this writing.

Question: What is the difference between Perplexity AI citations and ChatGPT citations?

Perplexity AI citations are displayed inline with footnote numbers that link to the source. ChatGPT citations appear as a list of sources below the response, with URLs included. Both engines prefer pages with named sources, short paragraphs, and direct answers. The format differs, but the citability factors are the same.

Question: How often should I update content to stay cited?

Update content when the facts change or when the client's product or service changes. Add a "Last updated" date at the top of the page. AI engines treat updated content as fresher and more reliable. If the content is evergreen and the facts have not changed, update it every 12 months to maintain the freshness signal.

Question: Can I test citability before publishing?

Run the citability checklist above on the draft. Check that the first paragraph answers the query in 80-200 characters. Check that every claim with evidence names a source. Check that schema markup is in place. If the page scores 70+, publish it. You cannot test whether it will be cited until it is live and indexed, but you can confirm that the structural elements are present.

Key Takeaways

  • AI engines cite content they can parse, verify, and attribute. Structure and named sources are the two biggest factors you control.
  • The first paragraph must answer the query directly in 80-200 characters using a definition verb.
  • Every claim that carries evidence must name the source. No "research shows" or "studies indicate" without an organization name.
  • Schema markup, author bylines, and last-updated dates signal freshness and authority.
  • Test citability by running the checklist above, then use the testing prompt to check whether the client's page appears in AI responses.

See It on a Client's Site

If you would rather see citability scored on a client's site than read about it, AISO Studio offers a free 7-dimension content audit at aiso.studio/audit. Run three audits with no account required. The audit scores fact-checking, structure, readability, SEO, WCAG accessibility, engagement, and GEO factors. It flags unsupported claims and missing citations.

For agencies running multiple clients, AISO Studio offers a 14-day full platform trial with no credit card required. The platform includes white-label PDF reports, client portal pages with your agency branding, one-click bundled scans, and direct WordPress publishing. Pricing starts at $99/month for freelancers managing 1-3 clients. Cancel any time from the dashboard. Start your trial at aiso.studio.

Ready to optimize your content for AI?

Get full agency-tier access for 14 days. Run audits, generate content, find leads — no credit card required.

Start Your 14-Day Pass