Does AI visibility need PR? What ChatGPT and Google AI Overviews actually cite

For the past year I kept hearing the same advice. If you want to show up in AI answers, you need PR. Get featured on external sites first and worry about your own website later.

I disagreed every time. In my trackers, the pages AI engines pulled from most often were on the brand’s own site: homepages, About pages, FAQs, product pages. But a few client trackers make an anecdote, so I decided to test the claim properly.

I used Zerorank AI to collect which sites and pages ChatGPT and Google AI Overviews cite in two niches: a local finance service and a B2B SaaS. Then I sorted every cited source by type and every cited page by format.

The numbers are clear. For both engines in both niches, content on brand websites carries most of the citations. Which content wins depends on the niche and on the engine.

How I collected the data

I tracked citations in two answer engines, ChatGPT and Google AI Overviews, for two kinds of business. I call the first one a local business: a local finance service. The second is a B2B SaaS.

Every cited source was classified as one of these types: own corporate site, competitor corporate site, editorial, UGC, institutional or other. Every cited page was classified by format, such as homepage, product page, listicle or how to guide.

I merged own and competitor corporate sites into one group called brand websites. Both are company sites, and for the engine the difference only matters to the company. In the text I still point out the split where it tells a story.

Who gets cited: brand websites lead everywhere

Here is the share of citations by source type. All numbers are percentages.

• ChatGPT leans hardest on company sites: 96% of its B2B SaaS citations and 79% of its local finance citations.
• Google AI Overviews are more mixed, but brand websites still account for 73% of local finance citations and 45% of B2B SaaS citations.
• Editorial and PR sources get 3% to 5% in three of four combinations. The exception is B2B SaaS in AI Overviews, at 16%.
• Institutional sources matter a little in local finance for ChatGPT (8%) and almost nowhere else.

The split inside brand websites is worth a look. In B2B SaaS in ChatGPT, competitors took 70% of citations and the brands’ own sites 26%. In local finance in ChatGPT it was 52% own and 27% competitors. Whoever has the better page for the question gets the citation. If your site has no good answer, a competitor’s site will fill the gap.

Which pages get cited: ChatGPT reads who you are, AI Overviews read how you answer

Source type is only half of the picture. The format of the cited page differs even more between engines.

ChatGPT: homepage first, product page second

In ChatGPT, the homepage took 65% of citations for the local finance service and 50% for the B2B SaaS. Product pages came next with 19% and 36%. Together those two page types cover 84% and 86% of the citations. Articles, comparisons and how to guides barely register.

I have seen the same thing in client work. When we added an FAQ and a clear “who we are” block to the Werty landing page and built out the About page, Google and the AI engines started pulling information from those pages as the main source. For 1tab, enabling server side rendering in April made the landing pages and About page readable for search engines and LLMs, and the English landing page conversion rate from search went from 9% to 19%.

Google AI Overviews: the format depends on the niche

For B2B SaaS, AI Overviews cited listicles (20%), comparisons (18%), articles (16%) and product pages (15%). The homepage got 1%.

For the local finance service the order changes. Product pages led with 27%, then articles (19%), how to guides (17%) and category pages (13%). Discussions made up 7% and the homepage 6%.

Put simply, ChatGPT tends to check who you are, and AI Overviews tend to look for a page that answers the exact question.

How often are listicles really cited?

Listicles are the format everyone argues about, so I compared my numbers with what other researchers report.

• An Evertune study of 25,000 cited URLs found that the major AI models share a heavy preference for listicles.
• Wix AI Search Lab data, summarized by AirOps, puts listicles at 21.9% of all AI citations, ahead of articles (16.7%) and product pages (13.7%). For SaaS the listicle share rises to 35.37%.
• XFunnel reports that listicles are 38% of the most cited pages across Google AI Mode and AI Overviews.
• Peec’s analysis of more than a million citations found that 52% of listicles reach ChatGPT’s top citation bucket, while Google AI Mode shows no format favorite.

My data agrees on one point and adds a limit. Listicles are a strong format for B2B SaaS in AI Overviews, where they made up 20% of citations next to comparisons at 18%. In ChatGPT they were only 4% for B2B SaaS and 2% for local finance, and they did not appear in the local finance AI Overviews at all. So the answer to “are listicles cited a lot?” is yes, in some places.

Two more cautions. Peec explains that listicles win because they hold extractable answers and many entities, which means the structure matters more than the label. And data shared by Peec’s Tomek Rudzki in August, as reported by 97th Floor, showed listicle citations in ChatGPT falling by about 50% around a model release. A format that works this quarter can lose its edge after the next update.

What about Reddit?

Reddit became the poster child of AI visibility advice: seed threads, get cited. My data and the public reporting both suggest a more careful reading.

Reddit is actively fighting AI spam

Reddit says its automated systems catch about 25,000 spammy posts and comments a day, and that users saw 20% less spam in the first quarter of 2026. It also revokes close to two million inauthentic votes a day, according to Engadget’s coverage. MediaPost describes the cause: brands and agencies flooding threads with stealth, AI written posts to get into chatbot answers. A tactic that platform staff are actively hunting has a short shelf life.

Reddit is cited less often than the hype suggests

Promptwatch data reported by Axios shows Reddit holding an average 3.83% of ChatGPT Search citations in late July and early August 2026, then falling to 0.52% in mid August. Over the same period its share dropped 11% in Google AI Overviews and 31% in Google AI Mode. Numbers differ by method: Semrush’s prompt set shows higher shares, while an agency that reviewed 39 monitored projects, iSocialWeb, saw Reddit among cited sources in only 6 of them.

My own numbers point the same way. UGC, the group that includes discussion platforms such as Reddit, got 0% of ChatGPT citations for B2B SaaS and 2% for local finance. In Google AI Overviews it jumped to 35% for B2B SaaS and 18% for local finance.

So Reddit and similar communities are an AI Overviews channel. They are a weak bet for ChatGPT. I have seen the AI Overviews effect myself: after we placed ten Reddit mentions for Werty and 1tab in April, Google’s AI Mode described Werty as often recommended, and it pulled that from our Reddit comments.

What to do with this

• Fix the homepage and the About page first. State who you are, what you do, for whom and where, with proof, pricing logic and an FAQ. ChatGPT cited the homepage in 50% to 65% of cases in this study. Make sure these pages render for crawlers.
• Give every service or product its own page with direct answers. Product pages were the second most cited type in ChatGPT and the top one for local finance in AI Overviews.
• B2B SaaS: add honest comparisons and listicles for AI Overviews. They made up 38% of AI Overviews citations in this niche. Keep them factual and updated.
• Local services: publish how to guides, articles and category pages. Together they were about half of AI Overviews citations for the local finance service.
• Join real conversations on discussion platforms for AI Overviews. Use real accounts and real expertise, and skip templated seeding.
• Use PR for what it does well. In my experience that means backlinks, brand mentions and local mentions that support map positions. Plan it after your own pages answer the questions.

Limits of this study

This is a comparison of two niches, two engines and one tracking tool at one point in time. Page type shares cover the most cited types only, so they do not add up to 100%. A citation also differs from influence: an engine can use a source in its answer without linking it. Treat the findings as direction, then check your own niche, your own prompts and your own competitors before you move budget.

Frequently asked questions