Which Sources Influence AI Answers About Brands? A 2,007-Company Study
Writesonic analysed 2,007 brands and reports an Influence Lift above 3× for YouTube, Reddit and Gartner. What the study supports—and what it does not prove.
A Writesonic study of 2,007 brands reports that a brand mention within a source cited by AI answers is associated with a higher probability that the brand appears in the answer. The largest published values among broadly represented sources are YouTube at 3.19×, Reddit at 3.17× and Gartner at 3.15×. The reported value for an owned website is 1.30×.
Those figures are useful only when read correctly. “Influence Lift” does not mean that one YouTube video or Reddit post mechanically triples brand visibility. The authors explicitly describe the results as directional rather than causal. The study indicates where a stronger adjusted association exists; it does not provide a guaranteed switch for appearing in ChatGPT, Gemini or Perplexity.
This analysis starts with Writesonic’s original study, published on 30 June 2026, and separates the measured result from commercial conclusions that the data cannot yet support.
The short verdict
The study supports three useful propositions. Citation frequency is not the same as the influence associated with a brand mention. Third-party sources containing discussion, comparison and evaluation can have a stronger relationship with brand presence than the owned website. Finally, effect size and the number of brands covered must be considered together.
It does not show that Reddit is “the new SEO”, that owned websites no longer matter, or that any paid mention creates visibility. An owned site remains the primary source for company-controlled facts, a citation destination and the verifiable foundation of brand identity. Independent sources provide confirmation and context that a company cannot credibly manufacture about itself.
What Writesonic measured
According to the published methodology, the researchers analysed more than 2,000 brands across major categories. For each AI-generated answer they identified cited pages, detected brand mentions within those pages and tracked whether the mentioned brand appeared in the answer text.
Influence Lift compares a brand’s probability of surfacing when a cited source mentions it with the probability when the source does not. Writesonic says its estimate isolates source effect from brand size, topic and query type. The public article does not, however, provide the complete statistical specification, confidence intervals or raw dataset required for full independent replication.
To enter the list of broadly applicable sources, a domain had to be cited across at least 600 brands. Low-quality automated statistics aggregators were excluded, and subdomains were normalised to their registered domain.
The principal values and their breadth
| Source or group | Reported Influence Lift | Published breadth |
|---|---|---|
| YouTube | 3.19× | 1,432 brands |
| 3.17× | 1,984 of 2,007 brands | |
| Gartner | 3.15× | 625 brands, concentrated in B2B |
| G2 / Capterra | 2.40× | professional review group |
| Amazon | 1.46× | marketplace |
| Wikipedia | 1.42× | widely cited source |
| Owned website | 1.30× | owned media |
Reddit’s lift is almost identical to YouTube’s while covering nearly the entire sample. Gartner reports a similar value across only 625 brands, concentrated in B2B. An ecommerce shop, an enterprise SaaS company and a consumer manufacturer therefore should not derive the same channel strategy from a similar number.
Citation frequency is not citation influence
Wikipedia may appear very frequently among domains cited by AI engines, yet the study assigns it a 1.42× lift for brand presence. Reddit may account for a smaller share of total citations while reaching 3.17× in this analysis. The measures answer different questions.
Frequency describes where systems retrieve or display information. Influence Lift attempts to estimate how much the probability of brand inclusion changes when a cited source mentions that brand. For resource allocation, the distinction matters: a heavily cited domain is not automatically the most valuable source for every company.
That is why our article on query fan-out and hidden queries separates the visible prompt from intermediate searches and displayed sources. The Writesonic study measures a relationship between citation, an on-page brand mention and answer presence; it does not expose every internal retrieval stage.
Why high lift on a thin sample can mislead
Writesonic notes that some niche domains produce lifts of 4× or 5× across only a small number of brands. A specialist running-shoe publication may be highly influential within that vertical and irrelevant to financial software.
Each opportunity requires two axes: lift and breadth. A large effect across 50 brands is a vertical signal. A roughly 3× effect across almost 2,000 brands may become a channel hypothesis. Neither guarantees an outcome for a specific company before its own prompt set is measured.
Why association does not prove causation
Strong brands tend to have more reviews, videos, discussions, budget and demand at the same time. Even after adjustment for size, topic and query type, unobserved variables may remain: awareness, product quality, age, distribution, PR activity and public interest.
A source mention may contribute to answer presence. It is also possible that already visible brands generate more source mentions. The relationship is likely bidirectional. Establishing causality would require controlled interventions, before-and-after series, comparable groups and greater transparency into the dataset.
The authors themselves list engine-level variation, recency and causality as open questions. Those qualifications belong beside the chart rather than hidden in a footnote.
The unit of analysis introduces another limitation. One page may mention several brands, while the same brand may occur across many pages from one source. Without the distribution of observations by brand, category and prompt, readers cannot assess concentration. An aggregate average may describe the dataset accurately while saying little about one narrow vertical. The published values are therefore useful for prioritising tests, not forecasting a specific company’s performance.
The owned website is the foundation, not the whole system
The 1.30× value does not mean a company should abandon its domain. Without precise official pages, AI products and third parties have nowhere reliable to verify specifications, pricing, policies, identity and updates. Owned content is factual infrastructure.
But a company website cannot independently validate claims such as “best”, “most trusted” or “preferred by customers”. Communities, publications, experts, professional reviews and comparisons address that gap. A healthy system connects both layers: the site publishes precise evidence while the external ecosystem tests, discusses and contextualises it.
Google recommends the same SEO fundamentals for its AI features: technical access, useful content, clear pages and accurate data. There is no special markup guaranteeing inclusion. Likewise, OpenAI explains OAI-SearchBot’s role in discovery, but crawler access is not a promise of citation.
What a Romanian brand should do
- Build the first-party base: product, pricing, terms, documentation, limitations and update dates.
- Map category sources: publications, communities, video channels, reviews, associations and authorities relevant to real questions.
- Measure before outreach: run a fixed prompt set and record displayed sources by engine, country and date.
- Contribute legitimately: original data, expertise, demonstrations, useful answers and product access—not concealed placements.
- Measure again: retain the same denominators and coverage instead of selecting favourable screenshots.
For a Romanian B2B service, Gartner may be inaccessible or irrelevant. Professional associations, economic media, partner documentation, conference videos and local technical communities may be the actual vertical sources. The map should emerge from citations observed in the category, not from copying a global ranking.
What not to do
Do not buy reviews, create fake Reddit accounts or turn communities into mention farms. Astroturfing can violate platform rules, damage reputation and produce signals that cannot survive public scrutiny.
Do not ask a publisher to insert a brand name into a semantically unrelated page. A useful mention must answer a real question and remain verifiable. The Writesonic study itself does not demonstrate that inserting a brand string is sufficient.
How to measure whether the strategy works
Use a before-and-after experiment with frozen prompts, replicates and the same observation window. Report presence, recommendation, citation, cited source, first-party versus third-party, mention prominence and accuracy separately. Our guide to measuring AI visibility explains why every rate needs a denominator and coverage.
Do not automatically attribute a change to the latest PR action. Engines change, sources update and generative answers vary. A defensible conclusion may be “visibility increased after the intervention within the observed set”, not “mention X caused the increase”.
Conclusion
The Writesonic study contributes an important distinction: the domains cited most frequently are not necessarily those most strongly associated with including a brand. In its sample, YouTube, Reddit and Gartner exceed 3× Influence Lift, while the owned site reaches 1.30×.
The mature interpretation is not “stop investing in your website and post on Reddit”. It is: maintain a verifiable first-party foundation, earn authentic validation within sources used by your category, and measure each engine separately. The study offers a map of hypotheses. Evidence for your own brand begins with your own controlled protocol.
Sources
- Writesonic — Which Sources Actually Influence AI Answers? 2,000+ Brands
- Search Engine Land — no universal top citation source for all brands
- Google Search Central — AI features and your website
- OpenAI — publishers and developers FAQ
The study and documentation were reviewed on 3 August 2026. Writesonic is a commercial GEO-tool provider; its figures are attributed and have not been independently replicated by Dosinescu.ro.