PR and brand teams: how press, encyclopedias, Q&A sites and reviews shape AI answers

This guide is for PR, communications and brand teams who are asked what their work means for AI answers. After reading it you can map the third-party sources that matter in your category and plan earned coverage for them. You will also know where the line runs between earning mentions and faking them. Plan on a few days to build the map, then fold the work into your normal PR calendar.

The short version

1. Map which third-party sites AI answers in your category actually cite.
2. Earn independent coverage with real news, data and access, not paid placements.
3. Help reviewers and list sites describe you accurately, and never buy rank or reviews.
4. Treat Wikipedia and Baidu Baike as rule-bound encyclopedias: qualify first, disclose, and cite independent sources.
5. Take part in Q&A communities openly, as a named representative.
6. Keep one set of facts consistent across every place that describes you.
7. Measure answers and cited sources before and after each campaign.

Map the sources answers cite

Start with evidence, not a media list. The sources AI answers draw on differ by assistant and category. Your own sample tells you where coverage would matter.

A 2025 study from the University of Toronto ran controlled queries across several verticals, languages and phrasings. It found that AI search engines showed a strong bias towards earned media (third-party, authoritative sources) over brand-owned and social content, unlike Google’s more balanced mix. In consumer electronics, about 93.7% of Claude’s and 93.6% of GPT’s sources were earned media.

Research-backed · Chen, Wang, Chen and Koudas (University of Toronto), arXiv:2509.08919

In DATA17’s August 2026 sample of one AI assistant (72 prompts, 297 citations), 80.5% of cited sources were third-party sites. The four most cited sources were all structured category lists (“best X tools” pages). Traditional media ranked far lower.

Observed in practice · observed 2026-08 · DATA17 AI Research Lab · 解数智能 (unpublished report, not publicly available; readers can't check it)

Our own re-test on consumer ChatGPT, in another category, found vendors’ own pages made up most citations (Claim: seeding brand mentions across the web boosts AI visibility). Chinese assistants also favour their own group’s content (Which sources Chinese AI assistants cite). So run your own baseline (Measuring AI visibility by sampling prompts):

• Ask 20–40 real customer questions in the assistants your customers use, with repeats.
• Record every cited domain and group it: news, trade press, review and list sites, encyclopedias, Q&A and forums, video, your own site.
• Rank the groups by how often they appear. That ranking is your earned-media target list.

Earn independent coverage

Reporters and analysts write about news, data and people. Those are what you can offer. A study, a customer outcome you can name, or an expert who answers questions on the record all give a journalist something to use.

The studies disagree on which kind of third-party content counts most, which is another reason to check your own category.

Meltwater reports that in more than 8 million AI citations it analysed, reviews and social proof made up 57% and thought-leadership content such as white papers 5.4%. In a smartphone case from its monitoring tool, brand websites were cited most, followed by YouTube (1,300 citations), tech media such as Tom’s Guide, TechRadar and CNET, and Reddit (542). The report does not say which period, prompts or assistants the figures cover.

Observed in practice · observed 2026-07 · Meltwater 融文 (unpublished report, not publicly available; readers can't check it)

DataStory asked the same questions on six Chinese AI platforms (web and app, 12 endpoints), three times each, in nine consumer categories (13–15 August 2026). Cited content was mostly ranking and recommendation lists (22.8%–72.0%), then product descriptions. User reviews were under 3% in every category.

Observed in practice · observed 2026-08 · 数说故事 (unpublished report, not publicly available; readers can't check it)

The two reports cover different markets and define content types differently. Neither shows that a press release on a wire service helps. Coverage that a journalist chose to write is what both kinds of study count.

An AI answer can’t cite coverage that doesn’t exist. Give journalists something worth writing, then let them write it.

Help reviewers and list sites get you right

List and comparison pages are among the most cited sources in several studies. You can’t control them, but you can make sure they have accurate material.

• Find the lists and review sites your baseline turned up. Check what they say about you.
• Send corrections with evidence: current prices, plans, features and limits.
• Offer trial access, test units and named customer contacts to reviewers.
• Ask happy customers to review you on the platforms they already use. Don’t pay for positive reviews or filter out negative ones.

The US FTC rule on consumer reviews bans writing, creating or selling fake reviews and testimonials. It also covers buying reviews, insider reviews, company-controlled review sites, review suppression and fake social media indicators.

Officially documented · US Federal Trade Commission (eCFR)

Encyclopedias: qualify first, then follow the rules

Wikipedia and Baidu Baike are widely read sources of entity facts. Both have strict rules about who may write about a company and from what sources. Breaking them gets edits reverted and accounts blocked.

Wikipedia presumes a company or product notable if it has been “the subject of significant coverage in multiple reliable secondary sources that are independent of the subject”. Press releases, press kits and coverage substantially based on them do not count.

Officially documented · Wikipedia

On Wikipedia, “COI editing is strongly discouraged”. Editors with a conflict of interest should propose changes on the article’s talk page rather than edit directly. Anyone paid to edit must disclose their employer, client and affiliation, which the Wikimedia Foundation’s terms of use require.

Officially documented · Wikipedia:Conflict of interest · Wikimedia Foundation Terms of Use

Baidu Baike says it only includes facts whose source can be found. Company websites, forums, blogs and self-media are restricted references. For a company’s honours, industry standing, founding date, funding and certifications, Baike requires reports from public media or records from business registration authorities.

Officially documented · 百度百科

What this means in practice:

• No independent coverage yet? Don’t start with an encyclopedia entry. It will be removed. Earn coverage first (previous sections).
• On Wikipedia, disclose your connection and post requests on the talk page with independent sources.
• On Baidu Baike, back each key fact with media reports or registry records. Your own site only supports basic facts such as products and address.
• Don’t hire anyone who offers guaranteed entries or “locked” content.

Encyclopedias reward the same thing AI answers do: facts someone else has checked. Earn the coverage, and the entry becomes possible.

Q&A sites and communities: show up as yourself

Zhihu, Reddit and similar communities are cited sources for some assistants (Which sources Chinese AI assistants cite). Answer real questions in your field from an account that says who you are. Link to evidence, admit limits and recommend competitors when they fit better. Don’t run networks of “independent” accounts. That is exactly what regulators and platform policies target (next section).

Keep your facts consistent

Coverage only helps if it says the same thing. Before each campaign, give spokespeople, agencies and partners one fact sheet: legal name, what you sell, founding year, locations, pricing model and key numbers, each with a source. Update it when anything changes, and correct outdated third-party descriptions. The fact page on your own site is covered in When AI gets your brand wrong: find it, fix it, prevent it.

What not to do

Google says its generative AI features “can show what’s being said about products and services across the web, including in blogs, videos, and forum discussions. However, seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem.”

Officially documented · Google Search Central

Google’s definition of spam includes “attempting to manipulate generative AI responses in Google Search.”

Officially documented · Google Search Central

The China AI Industry Alliance’s 2026 report lists the abuses it sees in GEO. They include forged expert identities and joint research, invented “authoritative” rankings, mass-produced fake user reviews, and fake negative news about competitors. They also include altering third-party encyclopedia and Q&A content to plant misleading facts.

Observed in practice · observed 2026-05 · 中国人工智能产业发展联盟（AIIA） (unpublished report, not publicly available; readers can't check it)

China’s 2026 Qinglang campaign lists AI data poisoning as a target, including forging authoritative data and using GEO for malicious marketing.

Officially documented · 新华网（中央网信办通知）

So avoid:

• Paid “advertorial” networks sold as AI optimisation. Mass-published near-identical articles are what the 2026 campaign targets (GEO regulation in China).
• Buying a place on “best of” lists without the list disclosing it is paid.
• Undisclosed encyclopedia editing, by your staff or an agency.
• Fake or incentivised reviews, and suppressing negative ones.
• Attacking competitors with planted negative content.

The test is simple: would the mention survive if everyone knew who arranged it? If not, it’s not earned media.

How you’ll know it’s working

• Same questions, same assistants, before and after. Re-run your baseline 4–8 weeks after a campaign lands (Measuring AI visibility by sampling prompts).
• Watch cited sources, not only mentions. Did the outlets, lists or encyclopedias you worked on start appearing as citations?
• Check accuracy. Are descriptions of you right, current and consistent? Errors are handled in When AI gets your brand wrong: find it, fix it, prevent it.
• Report per assistant. Sources differ by platform, so averages hide what changed.
• Expect lag. New coverage has to be published, crawled and picked up. A flat month is not a verdict.

Evidence at a glance

Recommendation	Evidence
AI search leans on earned (third-party) media	Research; observed in industry studies, but varies by assistant and category
Map cited sources in your own category first	Observed (studies disagree on which source types count most)
Encyclopedias need independent coverage and disclosure	Official (Wikipedia, Wikimedia, Baidu Baike rules)
Fake reviews and bought mentions are off limits	Official (FTC rule, Google spam policies, CAC campaign)
Seeking inauthentic mentions doesn’t help in Google	Official
Show up in communities as a named representative	Our recommendation, based on the policies above
Which source types matter most for AI answers	Observed only; reports disagree