Ask AI for the best option in your category and the model backs its picks with sources, but which kinds of sources differ sharply by industry. Across 5.17 million links between an AI recommendation and the source behind it in Parse's panel, finance comparison sites back a quarter of Financial Services recommendations and 1% of Software's, while retail marketplaces back a fifth of Commerce picks and almost none in finance. There is no universal citation source list. The source types that decide AI recommendations are category-specific. For the cross-category baseline of which source domains AI cites most, Parse has the underlying numbers.
- Across 5.17 million recommendation-to-source links (Parse first-party, October 2025 to June 2026, 11,130 prompts, 66,385 brands, 46 industries), the source-type mix behind AI recommendations is category-specific. Beyond a shared spine of the general web (34.9%) and the brand's own site (21.6%), the specialist layer diverges sharply.
- Finance comparison sites (Money.com, LendingTree, Experian, Credit Karma) back 24.8% of Financial Services recommendation evidence, 7× the 3.5% cross-industry baseline, and barely 1% of Software's.
- Retail marketplaces (Amazon, Walmart, Etsy, Target) back 19.7% of Commerce and 28.6% of Consumer Goods evidence, but under 1% in Financial Services and Software.
- Dedicated health sources (GoodRx, Verywell, Talkspace) back roughly 14% of Health Care and Biotechnology evidence, about 17× baseline, and effectively 0% in every other category.
- The brand's own site counts as evidence in B2B software (25% to 45% in Software, IT, and Apps) but is nearly invisible in consumer categories (1.6% in Consumer Goods, 0.4% in Clothing). Community sources (Reddit, YouTube) range from 7% in finance to 27% in Consumer Electronics.
Do the sources AI cites differ by industry?
Yes, and the gap is large enough to break a generic playbook. The instinct behind most AI visibility advice is that there is one list of sources to chase: get into the big directories, get cited on Reddit, fix your own pages. The data says the list is different in every category. We measured this on Parse's recommendation-evidence layer, which links each reviewed AI brand recommendation to the specific source the model cited for it. Across 5.17 million of those links, the source type that backs a recommendation swings from finance sites in one vertical to retail marketplaces in another to dedicated health publishers in a third. Two categories can look identical on a surface metric like citation count yet run on completely different evidence underneath. That means the first step in any AI visibility plan is not a tactic. It is finding out which kinds of sources actually decide recommendations in your category, because copying another industry's source list points your effort at the wrong places.
The shared spine every category runs on
Start with what does not vary, because it sets the baseline. Two source types show up everywhere. The general web (broad informational sites that are not clearly a marketplace, forum, or trade publication) backs 34.9% of recommendation links across all industries. The brand's own site, classified here as the SaaS or vendor source type, backs another 21.6%. Add community sources like Reddit and YouTube at 12.6% and you have roughly 69% of the evidence behind a typical AI recommendation coming from three generic buckets that exist in every category. This shared spine is why generic advice feels right: yes, your own pages and the open web matter everywhere. But the spine is also where everyone is already competing, and it is not where categories differ. The signal that tells you where to actually spend lives in the specialist layer stacked on top of it, and that layer is anything but uniform.
Which specialist sources back recommendations in each industry?
The specialist layer is the fingerprint. For each industry we took the source type it leans on most relative to the cross-industry baseline, the share of that industry's evidence it represents, and a few example domains. The table below covers a representative set of the 39 industries with at least 20,000 links; the over-index column is the industry share divided by the cross-industry baseline for that source type.
| Industry | Signature specialist source | Share of links | Over-index vs baseline | Example domains |
|---|---|---|---|---|
| Financial Services | Finance sites | 24.8% | 7.1× | Money.com, LendingTree, Experian |
| Consumer Goods | Retail marketplaces | 28.6% | 5.0× | Amazon, Walmart, Etsy |
| Health Care | Health publishers | 13.8% | 17× | GoodRx, Verywell, Talkspace |
| Biotechnology | Health publishers | 13.4% | 17× | health-focused sites |
| Education | Academic and education | 13.5% | 12× | university and edu sites |
| Travel and Tourism | Travel directories | 11.6% | 5.3× | Skyscanner, AAA Travel |
| Gaming | Trade and sports press | 20.5% | 4.2× | FoxSports, CBSSports, RotoWire |
| Consumer Electronics | Community (Reddit, YouTube) | 26.6% | 2.1× | YouTube, Reddit |
| Real Estate | Reference and directories | 9.2% | 3.2× | reference and listing sites |
| Blockchain | Developer and finance | 12.7% | 7.1× | dev docs and crypto-finance |
Read down the table and no two rows match. A retail marketplace is the dominant evidence in Consumer Goods and a rounding error in Financial Services. A finance site is the spine of finance recommendations and absent from health. The same "get cited" goal points at a different door in every category.
If you want to see how AI engines describe your own brand, run a free brand check — it takes a minute.
Finance, retail, and health each run on their own sources
The three sharpest examples are worth naming, because they show how far apart the recipes sit. In Financial Services, finance comparison and credit sites (Money.com, LendingTree, Insurify, Experian, Credit Karma) back 24.8% of recommendation evidence, 7× the 3.5% baseline, while retail marketplaces and health sites are near zero. In Commerce and Shopping, the picture inverts: retail marketplaces (Amazon, Walmart, Etsy, Target, Alibaba) back 19.7% of evidence, rising to 28.6% in Consumer Goods and 25.4% in Clothing and Apparel, while finance sites barely register. In Health Care and Biotechnology, dedicated health publishers (GoodRx, Verywell, Talkspace, and condition-specific organizations) back roughly 14% of evidence, about 17× the 0.8% baseline, and effectively 0% everywhere else. A finance brand that earns placements on retail marketplaces and a retailer that invests in finance publishers are each optimizing for the other's category.
Why software and AI look nothing like the rest
Tech categories are the clearest case of a different recipe, because their specialist layer is mostly missing. In Software, Information Technology, Artificial Intelligence, Data and Analytics, and Collaboration, the general web (46% to 52%) plus the vendor's own site (25% to 29%) plus community (10% to 13%) account for the large majority of evidence. The vertical specialist types that dominate other industries (finance, retail, health, directories) all sit near or below 1%. The only specialist signal that rises is developer and documentation sources, and even that is modest at 3.2% to 3.4%, roughly 2× baseline. The practical read is that AI visibility in software is won on the open web, your own documentation and marketing pages, and developer-leaning community, not on third-party verticals. It is also why your own site counts here: in Apps it backs 32.4% of evidence and in Events 44.7%, against 1.6% in Consumer Goods and 0.4% in Clothing. In consumer categories, the brand's own pages are nearly invisible as evidence.
Where Reddit and YouTube actually move recommendations
Community sources are often treated as a universal lever, but their weight is category-specific too. Across all industries, social and community sources (led by Reddit and YouTube) back 12.6% of recommendation evidence. The range around that average is wide. Community is heaviest in Consumer Electronics (26.6%), Sports (22.4%), Media and Entertainment (22.2%), and Clothing and Apparel (22.9%), where buyers lean on hands-on video reviews and forum threads. It is lightest in Biotechnology (6.9%), Financial Services (8.2%), and Professional Services (8.0%), where the evidence skews toward institutional and trade sources instead. In Consumer Electronics, YouTube alone supplies more than 12,700 of the links we examined and Reddit another 6,300. The lesson is not that community does or does not matter. It is that the same Reddit-and-YouTube push that is central in consumer electronics is a side bet in regulated verticals, and budget should follow that difference. For the community side specifically, see our analysis of which subreddits get cited most by ChatGPT and Parse's data on how YouTube ranks against Reddit as a cited source.
What a category-specific recipe means for your visibility plan
Treat your category's source mix as the map before you pick a route. The mistake this data exposes is importing a source list from a case study in a different industry, then wondering why the placements did not move anything. A finance brand should be auditing its presence on comparison and credit sites; a consumer-goods brand should be looking at marketplace listings and editorial roundups; a software brand should be looking at its own documentation, the open web, and developer community. Parse maps which domains and source types actually back recommendations in your category in Citations, so you can see your real recipe instead of guessing from a generic checklist. This pairs with a finding we have published before: there is no single directory chokepoint AI funnels through, as we showed in the directory tax barely exists. The two together say the same thing from opposite sides. The sources are diffuse, and the mix is local. Build for your category's recipe, not the industry's average.
How we measured this
The dataset is Parse's recommendation-evidence layer, one row per provenance link between a reviewed AI brand recommendation and the source it cited, joined to the source taxonomy and to industry through each prompt's primary niche. Scope is 5,174,354 links with an identified source type across 46 industry groups, 11,130 prompts, and 66,385 brands, over October 19, 2025 to June 19, 2026. Shares are computed over links with a non-null source type; unidentified-type links (about a quarter of the total) are excluded from the recipe. The source-type taxonomy is heuristically assigned and noisy at the margins, so the broad "general" and vendor-site buckets should be read as approximate while the specialist contrasts (finance, retail, health, directories) are the reliable signal. The unit is a repeated provenance link, not a unique user session, so a heavily cited domain weights its type. The window pools two collection eras; restricting to the current ChatGPT Search and Google AI Mode era reproduces the marquee contrasts (finance 27.9%, retail 20.2%, health 16.4%). Figures are aggregate and k-anonymous, and only industries with at least 20,000 links are reported.
Do AI models cite the same sources in every industry?
No. Across 5.17 million recommendation-to-source links in Parse's first-party panel, the source-type mix behind AI recommendations is category-specific. Every industry shares a generic spine of the general web (34.9%) and the brand's own site (21.6%), but the specialist layer on top diverges sharply: finance sites dominate finance, retail marketplaces dominate commerce, and dedicated health publishers dominate health care. A source type that backs a quarter of recommendations in one vertical can back under 1% in another.
Which sources does ChatGPT cite for finance versus software?
In Financial Services, finance comparison and credit sites (Money.com, LendingTree, Experian, Credit Karma) back 24.8% of recommendation evidence, about 7× the cross-industry average. In Software, those finance sites back only 1.3%. Software recommendations instead lean on the open web (about 47%), the vendor's own site and documentation (25% to 29%), and developer-leaning community. The same "get cited" goal points at completely different sources in the two categories.
Do directories or Reddit dominate AI citations?
Neither dominates universally. Directories back only 2.2% of recommendation evidence on average and peak around 11.6% in Travel and Tourism. Community sources like Reddit and YouTube back 12.6% on average but range from 7% in finance and biotech to 27% in Consumer Electronics. Both are category-specific levers, not universal ones, which is why a single tactic rarely transfers across industries.
How do I find which sources AI cites most in my category?
Look at the source mix behind recommendations in your specific niche rather than an industry average. Parse's Citations view maps which domains and source types back AI recommendations for your category, so you can see whether your category runs on marketplaces, finance sites, trade press, community, or your own documentation. That recipe tells you where placements are likely to move recommendations and where they will not.
Does the brand's own website count as a citation source?
It depends heavily on the category. The brand's own site backs 25% to 45% of recommendation evidence in B2B software categories like Software, Information Technology, and Apps, where documentation and product pages are primary references. In consumer categories it is nearly invisible: 1.6% in Consumer Goods and 0.4% in Clothing and Apparel, where AI leans on marketplaces, editorial roundups, and community reviews instead. Investing in your own pages pays off in software far more than in consumer goods.