What ChatGPT says about a brand is a reputation record that anyone can read in seconds. An AI answer compresses news coverage, reviews, forum threads and reference pages into a few sentences, and Pew Research Center found that Google users click fewer links when an AI summary appears. A brand can read its own paragraph, trace the sources behind it and change those sources.
Last updated: November 1, 2026. Data window for the 5W figures: September 27 to October 26, 2026.
Summary in six points
- Pew Research Center found that Google users clicked a traditional link in 8% of visits when an AI summary appeared and in 15% when none did.
- The AI paragraph is one part of a brand's reputation, and it is built from reviews, news, forum threads and reference pages.
- In 5W's data, earned media and social platforms supplied more citations than a single brand website.
- Different AI platforms name different brands for the same question, so one check on one platform is not enough.
- A wrong answer is changed by correcting or adding the pages the answer draws on.
- No brand can guarantee an AI answer, and the 5W figures are a company case study.
What changed when search results became one paragraph?
Search results changed from a list of links to a written answer, and the written answer now takes much of the attention. Pew Research Center analyzed the browsing of 900 U.S. adults in March 2025 and found that 18% of their Google searches produced an AI summary.
When a summary appeared, users clicked a traditional result link in 8% of visits. When none appeared, they clicked in 15% of visits. Users clicked a link inside the summary itself in 1% of visits.
Longer searches produced summaries more often, and just 8% of one- or two-word searches did. Questions about whether a company is worth hiring or buying from are usually written as full sentences, the kind of search Pew found more likely to produce a summary. Pew studied Google AI summaries only, not ChatGPT, and its full write-up is on the page about Google users and AI summary clicks.
Is the AI paragraph now a brand's reputation?
The AI paragraph is one part of a brand's reputation, and it is the part a buyer can read without visiting any other page. Reviews, court filings, news stories and forum threads still exist and still matter, because AI answers draw on them.
The paragraph is the compressed version of those inputs. An error in one source can appear in the answer as a single sentence that reads as settled fact, with no link beside it that a reader is likely to open.
This is the author's view, based on how buyers now ask questions, and it is not a measured share of buying decisions. None of the studies cited here measures how many buyers decide from an AI answer alone.
Which sources shape what AI says about a brand?
Outside sources shaped AI answers more than a single brand website did in 5W's data on PR agency questions. In the last 30 days, earned media sources earned 10.4% of citations and social and community platforms earned 8.0%, against 3.6% for 5wpr.com.
| Source type | Share of citations |
|---|---|
| Earned media sources | 10.4% |
| Social and community platforms | 8.0% |
| 5wpr.com (one site) | 3.6% |
| Institutions | 3.3% |
| PR wires | 1.1% |
Earned media and social platforms are categories made of many sites, while 5wpr.com is one site. The table shows where citations come from, not which single page matters most.
The publication matters as well as the story. On the same prompts, O'Dwyer's ranked eighth among all cited sources, PRovoke Media seventeenth, and PRWeek fell outside the top 25. A separate post on this site maps which websites AI answers cite most.
The source mix depends on the topic. 5W's data covers PR agency questions only, so a hospital, a bank or a consumer brand would see a different mix.
Why do different AI platforms write different paragraphs?
AI platforms write different paragraphs because each one selects different sources for the same question. Brand visibility measures how often a brand appears in AI answers to the tracked prompts, and 5W's data shows it moving by platform.
| Platform | 5W Public Relations | Edelman |
|---|---|---|
| Google AI Overviews | 37.6% | 18.5% |
| Perplexity | 33.2% | 25.6% |
| ChatGPT | 28.4% | 22.3% |
| Google Gemini | 26.7% | 18.2% |
| Google AI Mode | 24.6% | 18.5% |
| Microsoft Copilot | 22.4% | 28.3% |
5W had higher brand visibility than Edelman on five of six platforms, and Edelman led on Microsoft Copilot. The order also changed by topic. Edelman led on Healthcare PR (46.1% against 19.3%), while 5W led on 25 of the 31 tracked topics.
5W produced this analysis and is one of the two firms compared, so read the table as an illustration of how answers vary by platform and topic and not as a ranking of agencies.
How can a brand check what AI says about it?
A brand can check what AI says about it by running the same set of questions on several platforms and recording each answer. A log kept the same way each month shows what changed.
- Write ten questions that a buyer, a job candidate or a journalist would ask about the company.
- Run each question on ChatGPT, Google Gemini and Perplexity.
- Copy each answer into a sheet with the platform and the date.
- Record every source the answer cites and every brand it names.
- Mark each claim in the answer as accurate, outdated, wrong or missing.
- Repeat the same questions every month.
Platforms can return different wording each time a question is run, so compare one month against the next and not one run against another. Tracking which domains are cited across a fixed set of questions is the basis of citation share, and a separate post explains how AI engines decide what to cite.
How can a brand change a wrong or outdated AI answer?
A brand changes a wrong or outdated AI answer by correcting or adding the pages the answer draws on, because the answer is written from retrieved sources. Each action below follows from that mechanism.
- If a cited page contains the error, ask that publisher for a correction, because the answer repeats what the retrieved page says.
- If no page states the correct fact, publish a dated page on your own site that states it in the wording of the question, so retrieval has a page to select.
- If the answer leaves out a real strength, publish evidence for it with a named method, such as a count, a survey or an audit.
- If a forum thread or review is the source, reply on that platform with the facts and the date, because the thread stays retrievable.
The case for publishing evidence comes from a study by researchers at Princeton University, IIT Delhi and other institutions. In a benchmark of 10,000 queries, adding citations, quotations from relevant sources and statistics raised source visibility by more than 40% across queries. The result varied by topic and came from an experiment, so it describes a tendency and not a promised outcome for any brand. The paper is available as GEO: Generative Engine Optimization on arXiv.
This work sits where reputation management meets generative engine optimization, and it is part of AI communications.
What should a brand not expect from this work?
A brand should not expect a guaranteed answer, because AI platforms choose their own sources and change how they choose them. Results vary by brand, category, competitors and platform.
Three terms are easy to confuse. A citation means an AI answer used a page as a source. A mention means the answer names the brand. A recommendation means the answer presents the brand as a good fit, and the 5W data in this article does not measure recommendations.
The first step costs nothing. Write the ten questions a buyer, candidate or journalist asks about your company, run them on ChatGPT, Gemini and Perplexity this week, and trace the first wrong sentence back to the page it came from.
How was the data in this article collected?
The 5W figures come from 5W's analysis using Profound's AI visibility tools for September 27 to October 26, 2026. 5W produced the analysis and uses Profound as a customer, so it is a company case study and not independent validation, and it is not a Profound study.
Method and disclosure
5W selected all 396 prompts, and no outside party reviewed them. The prompts cover PR agency questions and ran in two batches. A total of 190 began January 27, 2026 in the US, UK and Canada on ChatGPT, Copilot, Perplexity and Google AI Overviews. The other 206 began September 14 in the US only, on those four platforms plus Gemini and Google AI Mode. The batches differ in topics, regions, platforms and run length, so totals should not be compared across them. The full set of tables is in 5W's chatbox report.
The Pew figures come from Pew Research Center's analysis published July 22, 2025, which used March 2025 browsing data from 900 U.S. adults. The 40% figure comes from the paper by Pranjal Aggarwal and co-authors, published at the ACM SIGKDD Conference in August 2024.
Frequently asked questions
What does ChatGPT say about a brand?
ChatGPT's answer about a brand depends on the wording of the question, the platform and the date. The same question can return different sources and different wording, so a brand should run the same questions each month and compare the results.
Can a brand edit what ChatGPT says about it?
No. A brand cannot edit an AI answer directly. It can correct the pages the answer draws on, publish dated pages that state the correct facts and add evidence, which changes what retrieval has to choose from.
Does an AI answer matter more than a Google ranking?
The data in this article does not rank the two. Pew found that users clicked a traditional link in 8% of visits when a Google AI summary appeared and in 15% when none did, so both the answer and the ranking affect what a reader sees.
How often should a brand check its AI answers?
Monthly is a workable cycle. In 5W's data, one page earned 52% of its lifetime citations in the last 30 days while another earned 6%, so citation patterns can shift within a month.
Is the 5W data independent?
No. 5W produced the analysis, chose all 396 prompts and uses Profound's tools as a customer. No outside party reviewed the prompt list, and results for any other brand will vary.