GrowNexus — Your Growth, Our Strategy
ServicesIndustriesMarketsProcessWorkInsights
GrowNexus — Your Growth, Our Strategy

360° growth partner for Kathmandu, Birgunj and global markets.

  • Facebook
  • Instagram
  • LinkedIn
  • TikTok

Digital Growth

  • Branding & Creative
  • Design Services
  • SEO Optimization
  • Growth Strategy & Consulting
  • Business Automation & AI
  • Training & Certification
  • Social Media Marketing
  • Website & App Development

Activation & BTL

  • Activation & Experiential
  • Event Marketing
  • Retail & Shopper Marketing
  • Direct Marketing
  • Influencer & Community
  • Promotions & Campaigns
  • Trade Marketing (B2B)
  • Out-of-Home (OOH)
  • Print & Collateral
  • Corporate Gifting & Merch
  • PR & Brand Building

Industries

  • Hospitality
  • E-commerce
  • SaaS
  • Professional Services
  • Education
  • Real Estate
  • PR & Communications
  • News & Digital Publishers

Markets

  • Markets hub
  • Nepal
  • United States
  • United Kingdom
  • Australia
  • Europe
  • UAE
  • India

Resources

  • Insights
  • Process
  • Pricing approach
  • Contact

Growth insights, no fluff

Monthly notes on SEO, paid social, and AI-native marketing for teams that ship.

© 2026 GrowNexus. All rights reserved.

  • Privacy
  • Sitemap
  • Terms
  • Cookies
  • Accessibility Statement
Skip to content
GrowNexus — Your Growth, Our Strategy
WorkInsightsAbout
Book a strategy call
AI & Automation
August 22, 2026·8 min read

How to Format a Press Release So AI Assistants Cite It

One checkable fact, a named human, real questions, and the right schema. What actually makes a release extractable in 2026 — including the FAQ markup change most advice has not caught up with.

A press release document with seven blocks. Three of them are highlighted and trace dashed lines out of the document into a separate answer panel; the rest stay put.

Article body

On this page
  • Lead with the fact, and state it once
  • The quote is the citation unit
  • Answer the question in the buyer's words
  • Schema: what to mark up, and what stopped paying
  • Where the release lives changes what it can do
  • The boilerplate is an entity record, not a paragraph
  • What does not work
  • The checklist

On this page

  • Lead with the fact, and state it once
  • The quote is the citation unit
  • Answer the question in the buyer's words
  • Schema: what to mark up, and what stopped paying
  • Where the release lives changes what it can do
  • The boilerplate is an entity record, not a paragraph
  • What does not work
  • The checklist

Short answer: a citable release states one checkable fact near the top in plain language, attributes it to a named person with a resolvable profile, answers the question in the words buyers actually use, and carries Organization, Person and NewsArticle markup so the entity is unambiguous. Formatting alone will not earn a citation, but bad formatting reliably prevents one.

Most advice about press releases and AI stops at "add schema." That is the smallest part of it. An answer engine has to find the release, decide it is relevant, choose it over the twenty other sources covering the same news, and then have something in it clean enough to quote. Structure only helps at the last two steps — but that is where most releases fail, because they are written to satisfy an editor's format rather than to be extracted.

This is about the release itself. For what to do once it is published and how to tell whether any of it worked, see measuring AI citations from PR. For the broader evidence on what gets any page cited, see our analysis of 18,151 cited sources.

Lead with the fact, and state it once

The inverted pyramid was designed for exactly this and then got buried under marketing language.

An extractable opening does three things: it names the entity, states what changed, and attaches a number or a date that can be checked. "Acme has raised a 12 million dollar Series A led by Northwind Capital, bringing total funding to 18 million" is a sentence an assistant can lift whole. "Acme, a leading innovator in workflow intelligence, is thrilled to announce a significant milestone" contains no facts at all.

Two rules follow:

  • State the central fact once, in one sentence, near the top. Repeating it in three variations across the release gives a model three slightly different candidates and no signal about which is canonical.
  • Put the number in the sentence, not in a chart. Anything that lives only in an image is invisible.

The quote is the citation unit

Quotations were one of the three strongest tactics in GEO: Generative Engine Optimization (Aggarwal et al., KDD 2024), the most useful controlled study available on this — alongside adding statistics and citing sources, at roughly 30 to 40 percent relative visibility gain on the paper's metric. Notably, lower-authority domains gained the most.

A quote earns that only if it carries information. Compare:

"We are excited to partner with Northwind on this journey."

"Enterprise buyers were taking 90 days to approve a purchase they had already decided on, so we rebuilt the approval flow first."

The second is quotable because it says something. It also does the other job a quote does: it attaches a named human to a claim.

Name the person properly. Full name, exact job title, and a link to a profile page that actually exists on your domain. A model resolving "who said this" needs somewhere to land. A quote attributed to "a spokesperson" is unattributable by design.

Answer the question in the buyer's words

Add a short question-and-answer block using the phrasing people actually type. Not "About the product" — the literal questions.

Get the wording from real sources: your own Search Console conversational queries, sales call recordings, and the questions competitors answer on their comparison pages. Invented questions read as invented.

Three to five is plenty. Each answer should stand alone without the surrounding release, because that is how it will be read.

Schema: what to mark up, and what stopped paying

This is where the advice on the open web is most out of date.

MarkupStatus in 2026What it does for you now
OrganizationSupportedResolves who published it; the foundation of entity clarity
Person (on the quoted spokesperson)SupportedConnects a claim to a named human with a profile
NewsArticle / ArticleSupportedIdentifies the content type, publish date and author
speakableStill supported, but limited to news-like content and English-heavy coverageFlags the passage best suited to be read aloud or extracted
FAQPageRich results deprecatedNo search appearance. Still valid schema.org, still useful for machine extraction

The FAQPage change is the one that catches people out. Google narrowed FAQ rich results to authoritative government and health sites in August 2023, and in June 2026 removed the search appearance, the rich result report, and FAQ support from the Rich Results Test entirely — ending eligibility even for the sites that had kept it (Search Engine Journal).

Keep the markup, change your reason for using it. Google has said unused structured data causes no problem, FAQPage remains a valid type, and it still gives an extraction system a clean question-and-answer structure. What it will never again do is win you a bigger blue link. If the only reason your release has an FAQ block is the rich result, delete the block. If the reason is that a buyer or an assistant will read it, keep it and write it properly.

Validate with the Schema Markup Validator rather than Google's Rich Results Test, which no longer reports on FAQ at all.

Where the release lives changes what it can do

A release that exists only on a distribution network is a page you do not control, cannot update, and cannot mark up. Before anything is distributed, it should exist on a URL you own.

Publish to your own newsroom first

Give every release a permanent URL on your domain, published before or at the same moment the distribution goes out. That page is the one you can add schema to, correct if a fact changes, link internally, and point a profile at. It is also the version that carries your Organization markup rather than the network's.

Set the canonical on your own page to itself. You are not trying to consolidate signals from mirrors — you are making sure the version you control is unambiguous about being the original.

Make the newsroom itself crawlable

An index page listing releases in reverse-chronological order, linked from your main navigation, with clean URLs and no infinite-scroll-only pagination. This sounds obvious and is one of the most common failures we see: newsrooms built as a JavaScript widget that renders nothing to a crawler, or paginated only through a "load more" button that no bot will press.

Check what the page actually returns without JavaScript. If your releases are not in the HTML, they are not in the corpus.

Date everything, visibly

Publication date on the page, in the markup, and in a format a human can read. Recency is a live signal for news content, and an undated release is treated as an undated release. If you update a release with a correction, say so and stamp it.

The boilerplate is an entity record, not a paragraph

The "About" block at the foot of a release is usually the least edited text a company owns, which makes it the most repeated description of that company on the internet. Answer engines notice consistency.

Make it factual and keep it identical everywhere: legal name, what the company does in plain words, where it operates, founding year, and a link to the canonical site. Then make sure your site, your profiles and your releases all say the same thing. Contradictory boilerplate is a slow, invisible tax on entity resolution.

What does not work

Optimised anchor text in the release body. Google's spam policies name this explicitly, listing "links with optimized anchor text in articles, guest posts, or press releases distributed on other sites" as link spam. Link to your homepage and your newsroom with plain, descriptive text.

Keyword stuffing. In the GEO study it performed worse than doing nothing.

Volume as a strategy. Publishing the same release to hundreds of mirror domains produces hundreds of near-identical pages. That is a deduplication problem for a retrieval system, not an authority signal.

Gating the detail. Anything behind a form is invisible to the systems you are trying to reach.

The checklist

  1. One checkable fact, stated once, in the first two sentences.
  2. A named spokesperson with an exact title and a working profile link.
  3. At least one quote that contains information rather than enthusiasm.
  4. Three to five real questions with self-contained answers.
  5. Organization, Person and NewsArticle markup; speakable if it is genuinely news.
  6. A boilerplate identical to the one on every other property you own.
  7. Plain anchor text.
  8. Every number sourced and dated in the text, not in an image.

None of this guarantees a citation, and anyone telling you otherwise is guessing. What it does is remove the reasons a system would pass over you in favour of a competitor covering the same story — which, when the news itself is comparable, is usually what decides it.

We do this work with PR agencies, communications teams and distribution platforms, alongside the search and answer-engine programs and PR and brand building it supports. If you want the release template we use, get in touch.

Cite this article

Arman Ahamed (2026). "How to Format a Press Release So AI Assistants Cite It." GrowNexus, August 22, 2026. https://grow-nexus.com/insights/press-release-ai-citation-format/

Share on LinkedIn

Written by

Arman Ahamed

Digital Marketing Manager, GrowNexus

Leads growth strategy and AI-native marketing programs for Nepal and global clients. Writes about what ships, not what slides.

Who this is for

  • PR & Communications marketing

Keep reading

Related insights

More from the same topic, selected by category overlap.

AI & Automation
Aug 22, 2026/10 min read

How to Measure AI Citations from PR When Your Analytics Show Nothing

Assistant referrals look like a rounding error while page-one buying questions earn no clicks. Four measurement methods, what each one misses, and the three report splits that change what you work on next.

AI & Automation
Aug 13, 2026/12 min read

What Gets a Business Cited by AI Assistants? 18,151 Sources Compared

We re-analysed 18,151 cited pages across ChatGPT, Google and Perplexity. Relevance, structure and evidence were associated with deeper use, but they are not magic ranking factors.

AI & Automation
Aug 13, 2026/8 min read

How to Write an llms.txt: A Guide to v2, With 25 Sites Measured

The llms.txt spec was revised in August 2026. We measured 25 major sites on 13 August: 18 publish a file, but only 2 have implemented the discoverability relations v2 added, and none ship them in HTML. Here is how to write one properly.

Strategy notes, no fluff

One email when we publish something worth your time. Unsubscribe anytime.