Key findings
- Moz appeared in 78.8% of SEO answers, 19.4% of marketing answers, 5.2% of AI answers and 2.5% of answers outside the four named topic groups.
- The six most-named SEO sources were Moz, Semrush, Search Engine Journal, Ahrefs, Search Engine Land and Backlinko. Each appeared in at least 52% of SEO answers.
- Forbes led the AI-topic answers at 43.1%. OpenAI appeared in 10.3% of them.
- Mean cross-model Jaccard overlap was 0.204 for VPN and security, 0.197 for SEO and 0.071 for AI, so the models’ source lists overlapped more in the first two groups.
The unit here is an answer that names a source. A name in a generated list isn’t automatically a recommendation or a live citation, and that difference matters when you’re reading brand mentions and citations side by side.
The overall averages hide how different each topic is
Across all 14,205 answers, the five most-named sources were four broad publishers and one software brand:
| Rank | Source | Share of all answers |
|---|---|---|
| 1 | Forbes | 25.6% |
| 2 | TechRadar | 17.2% |
| 3 | HubSpot | 15.4% |
| 4 | Harvard Business Review | 14.3% |
| 5 | TechCrunch | 14.2% |
Forbes leads the pooled sample, but split the answers by topic and the picture flips. Semrush appeared in 11.0% of all answers and 78.5% of SEO answers. Moz sat at 9.8% overall and 78.8% in SEO. An overall share mostly tells you which topics are in the sample.
The answers contained 324,131 source mentions and 79,218 distinct cleaned source names. Only 3,680 of those names (4.6%) appeared in ten or more answers, and 72.9% appeared exactly once. The top ten names accounted for 6.2% of mentions and the top hundred for 22.7%, so the long tail is very long.
SEO answers went to SEO specialists. VPN answers didn’t.
The four named topic groups contained:
- SEO: 1,040 answers, including keywords such as “seo agency” and “seo for ai agents”.
- VPN and security: 1,631 answers, including “best vpn for disney plus” and “best vpn for streaming iptv”.
- Marketing: 1,050 answers, including “digital marketing and ai”.
- AI: 3,705 answers, including “ai consultant”, “generative ai tools” and “best ai tools”.
The remaining 6,779 answers fell into an “Other” group. Topics were assigned by keyword-matching rules, covered in the method section.
| Topic | Six most-named sources (share of answers in that topic) |
|---|---|
| SEO | Moz 78.8%, Semrush 78.5%, Search Engine Journal 73.1%, Ahrefs 72.1%, Search Engine Land 60.4%, Backlinko 52.1% |
| VPN & security | TechRadar 63.4%, CNET 57.7%, PCMag 57.3%, ExpressVPN 54.8%, NordVPN 54.6%, Tom’s Guide 42.7% |
| Marketing | HubSpot 63.8%, Forbes 42.2%, Sprout Social 36.9%, Hootsuite 36.7%, Buffer 34.5%, Neil Patel 29.0% |
| AI | Forbes 43.1%, Harvard Business Review 32.3%, TechCrunch 30.7%, McKinsey & Company 22.5%, HubSpot 21.4%, Gartner 21.3% |
Every SEO top-six source is an SEO tool or publication. VPN looks different: tech review publishers hold four of the six spots, ExpressVPN and NordVPN take the other two, and TechRadar at 63.4% shows up more often than either VPN brand.
So “specialists win” is too simple. A source can own a topic as the provider, as a specialist publication or as a broad reviewer, and CNET’s strong showing in VPN answers doesn’t make it a VPN specialist.
Moz’s share drops sharply outside SEO
Moz's share of SEO-topic and AI-topic answers in this sample
A gap of 73.6 percentage points, measured across two different pools of keywords.
Source: Wellows keyword analyser; 1,040 SEO answers and 3,705 AI answers| Source | Reference topic and share | Marketing answers | AI answers | Other answers |
|---|---|---|---|---|
| Moz | SEO 78.8% | 19.4% | 5.2% | 2.5% |
| Ahrefs | SEO 72.1% | 18.4% | 5.3% | 2.3% |
| Neil Patel | SEO 51.1% | 29.0% | 4.7% | 2.0% |
| CNET | VPN 57.7% | 2.8% | 6.9% | 7.8% |
| NordVPN | VPN 54.6% | 0.0% | 0.1% | 0.7% |
“Other” is everything outside SEO, VPN and security, marketing and AI, from “cat shoes safety” to “best dell laptop” to “medical detox ontario”. It’s a mixed remainder rather than one distant topic.
Moz, Ahrefs and Neil Patel all held up better in marketing answers than in AI or Other. HubSpot runs the opposite way from the SEO tools: 63.8% in marketing, 48.7% in SEO, 21.4% in AI and zero appearances in VPN answers.
What these numbers describe is topic-specific association. They don’t measure how much content each source has published, so they sit next to the topical authority question rather than settling it.
Broad publishers led the AI answers
Share of AI-topic answers that named Forbes
Harvard Business Review followed at 32.3% and TechCrunch at 30.7%.
Source: Wellows keyword analyser; 3,705 AI-topic answersThe AI top six is business and tech publications, consulting and research firms, and HubSpot. OpenAI turned up in 10.3% of AI answers, and slightly more often (12.8%) in marketing answers.
Part of that is the bucket itself. The AI group mixes broad commercial and informational keywords, while the SEO group covers one specific discipline, so broad publishers have more room to win.
If you’re working in this space, look at the actual buyer questions behind the aggregate. A publisher that leads broad AI questions may barely appear on the specific product category or technical subtopic you care about.
Model overlap was highest for VPN and SEO, lowest for AI
For keywords answered by more than one model, we compared the sets of named sources using Jaccard similarity: the number of sources both models named, divided by the number either one named. Zero means no shared sources, one means identical lists.
| Topic | Keyword–model-pair comparisons | Average overlap | Median overlap |
|---|---|---|---|
| VPN & security | 992 | 0.204 | 0.114 |
| SEO | 789 | 0.197 | 0.167 |
| Other | 4,619 | 0.122 | 0.075 |
| Marketing | 742 | 0.115 | 0.058 |
| AI | 2,481 | 0.071 | 0.014 |
Mean overlap in VPN and security was about 2.9 times the AI value, and SEO about 2.8 times. The medians make the gap starker: 0.167 for SEO against 0.014 for AI.
Each row counts keyword–model-pair comparisons, so a keyword answered by more models contributes more pairs. Overlap also moves with list length and with which model pairs are being compared.
If you want the cited-source version of this question, our AI citation overlap study covers it with a different dataset and setup.
Gemini names the big publishers least often
We counted answers naming at least one of Forbes, Harvard Business Review, TechCrunch, McKinsey, Wired, The Verge, Gartner or Deloitte, a fixed group of eight broad publications and firms.
| Model | Answers | Sources named per answer | Answers naming at least one selected broad source |
|---|---|---|---|
| Claude | 2,555 | 30.7 | 39.7% |
| Llama | 994 | 23.2 | 36.8% |
| DeepSeek | 1,251 | 18.0 | 35.7% |
| GPT | 8,319 | 22.0 | 33.5% |
| Gemini | 1,086 | 15.5 | 20.3% |
Gemini came in at 20.3%, against 33.5% to 39.7% for the other model families. It also returned the shortest source lists, at 15.5 names per answer, and shorter lists leave fewer chances to include one of the eight.
How to use this in your own visibility tracking
The clearest lesson is to look at visibility by topic. Moz’s 9.8% overall share and 78.8% SEO share come from the same dataset, and they tell very different stories.
- Choose questions that represent your buyers. Separate your core product category from adjacent subjects and broad industry questions, and keep each group small enough to read through by hand.
- Compare the same prompts across engines. Record the date, engine or model version, location and search settings, or a change in your test will look like a change in visibility.
- Report mentions and citations separately. A named source and a linked source answer different questions about your presence.
- Keep the denominator visible. Show how many questions and answers each topic contains next to the share that names your brand.
- Test expansion instead of assuming it. A low share in an adjacent topic tells you where to look. It doesn’t tell you which publishing or outreach move will change it.
Our companion study of citation coverage comes at the same question from the citation side, testing whether citations on one set of questions line up with citations on held-out questions in the same topic.
Wellows tracks where brands are mentioned and cited across ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode. All five engines are included on every plan, with a 7-day free trial.
Method
- Dataset: 14,205 usable answers from Wellows’ keyword analyser, covering 7,174 keywords from 901 users. Answers returning an error or refusal were excluded.
- Models: GPT, Claude, Gemini, DeepSeek and Llama. Not every keyword was answered by all five. GPT supplied 8,319 answers (58.6% of the sample), and the average was about 1.98 answers per keyword.
- Measurement: the websites, publications and other sources named in model answers.
- Sampling: keywords were chosen by Wellows users, so the topic mix reflects what they track.
- Share of answers: the percentage of answers in a group that named a source at least once. Repeating a source inside one answer doesn’t raise its share.
- Topics: keyword matching applied in the order SEO, VPN and security, marketing, then AI. That’s why “seo for ai agents” sits under SEO and “digital marketing and ai” under marketing. The 6,779 unmatched answers (47.7%) form Other.
- Source types: labels such as SEO-focused source, VPN provider or broad publisher describe the source’s role.
- Overlap: Jaccard similarity between model responses to the same keyword.
- Name cleaning: names were lowercased, with protocol prefixes, “www.” and the endings .com, .org, .net and .io removed. Variants such as “HubSpot Blog” stay separate from “HubSpot”.
