What AI Engines Cite Changes Completely By Industry
We ran the same measurement for eleven companies in eleven different industries, 1,169 tracked questions put to five answer engines. Directories and marketplaces were cited in 352 of 641 answers for a coworking operator, 54.9%, and in 11 of 250 answers for a healthcare benefits company selling to clinics, 4.4%. Same method, same engines, same week.
Most published AI citation statistics quote one number for the whole web. This page is the argument against that. Eleven measurement runs, one per company, taken between 21 and 23 October 2026, produced 5,659 answers and 28,598 cited links across 10,002 distinct domains. One thing held everywhere: vendor and agency websites were the most cited kind in all eleven industries, within a band of 317 of 500 answers, 63.4%, to 461 of 514 answers, 89.7%. Everything underneath that moved enormously. Government and regulator sites were cited in 69 of 255 answers, 27.1%, for a healthcare benefits company selling to employers, and in 1 of 644 answers, 0.2%, for an ecommerce analytics software company. That is the same source kind, measured the same way, 174 times apart. If you are deciding where to spend to be cited, the industry you are in changes the answer more than anything else on this page. B2B AI citation rate by industry turns these numbers into what to do from a low baseline.
Vendor sites vary 1.4 times across eleven industries. Government and regulator sources vary 174 times.
Refreshed, 27 October 2026
This page was first published on 23 September from runs dated 17 to 22 September. It now reports runs dated 21 to 23 October, and five things changed.
Eleven accounts, not ten. Two accounts on the September version are no longer measured, so software supply chain security and financial market data drop out. Three come in: a local restaurant, an ecommerce analytics software company and a work management software consultancy.
Buyer questions only. Each account's branded comparison questions are no longer counted, which settles the mixing our 29 September correction flagged.
One answer per question per engine. Four accounts are now measured from several countries. Where the same question went to the same engine from several countries, only the US answer is counted.
No cap on sources. The six-source limit described in the 29 September correction no longer applies, so the link counts below are complete, not floors.
Google Maps counts as a listing. Links to google.com/maps are now counted under directories and marketplaces rather than docs and standards, because they point at a business listing, not at documentation.
Why these dates: the runs after 23 October came back without any ChatGPT source lists we could capture, so each account's figures come from its most recent run in which all five engines returned sources.
Exactly what was measured
Every number on this page is computed in SQL against eleven stored snapshots. The SQL is printed in full at the end of the page source so anyone can reproduce it.
| Parameter | Value |
|---|---|
| Runs | 11, for each of 11 accounts the most recent run with sources from all five engines, one account per industry |
| Run dates | 21 to 23 October 2026 |
| Questions | 1,169 unbranded buyer questions in total, each account has its own tracked set of 50 to 151. Branded comparison questions are not counted |
| Answer engines | 5: ChatGPT, Gemini, Microsoft Copilot, Google AI Mode, Google AI Overviews. The ecommerce development account is measured on four and has no Google AI Overviews answers |
| Location | Where the same question went to the same engine from several countries, only the US answer is counted, so each question counts once per engine |
| Engine cells | 5,717. One question on one engine. 21 expected cells were absent from their snapshot |
| Answers excluded as no-answer | 58 of 5,717, 1.0%. An engine that returned no answer text cannot have cited anything |
| Answers counted | 5,659. This is the denominator for every share on this page |
| Answers with no source list | 758 of 5,659, 13.4%. Kept in the denominator, reported in section 06 |
| Cited links counted | 28,598 |
| Link exclusions | 1,332 links removed: Google search and result-viewer links and Bing redirect links, which point at the engine's own result panel rather than at a source. 839 were in the restaurant run and 350 in the coworking run |
| Google Maps | 163 links to google.com/maps, counted as one source, google.com/maps, under directories and marketplaces |
| Distinct domains | 10,002, counted at the registrable domain so subdomains roll up |
| Unit of the share | an answer, meaning one question on one engine. An answer counts once for a source kind however many links of that kind it cited |
Three of the eleven accounts are our own and are named: DealSource Systems, Danish Lead Co and SourceRank AI. The eight client accounts are shown as an industry label only. No account's own visibility score, pricing or commercial volumes appear anywhere on this page.
Each cited URL is reduced to a registrable domain and assigned to exactly one of seven source kinds by an explicit named rule set, printed in the SQL at the end of the page source. Vendor or agency site is the residual, which means it also contains each measured company's own website and its direct competitors. This is a companion to the single-account citation report, which takes one of these eleven runs apart domain by domain.
Source mix by industry
Share of that industry's answers citing at least one source of each kind. Rows are ordered by the directories and marketplaces share. An answer can cite several kinds, so rows do not sum to 100%.
| Industry | Vendor or agency site | Directories and marketplaces | Docs and standards | Government and regulator | Social and video | Community | News and media |
|---|---|---|---|---|---|---|---|
| Coworking and flexible office space account not named | 501 of 641 78.2% | 352 of 641 54.9% | 0 of 641 0.0% | 29 of 641 4.5% | 11 of 641 1.7% | 4 of 641 0.6% | 1 of 641 0.2% |
| Local restaurant account not named | 317 of 500 63.4% | 229 of 500 45.8% | 2 of 500 0.4% | 13 of 500 2.6% | 29 of 500 5.8% | 10 of 500 2.0% | 0 of 500 0.0% |
| Construction and industrial staffing account not named | 370 of 449 82.4% | 74 of 449 16.5% | 16 of 449 3.6% | 65 of 449 14.5% | 35 of 449 7.8% | 4 of 449 0.9% | 9 of 449 2.0% |
| Private equity and M&A deal sourcing DealSource Systems | 602 of 687 87.6% | 100 of 687 14.6% | 29 of 687 4.2% | 20 of 687 2.9% | 93 of 687 13.5% | 22 of 687 3.2% | 18 of 687 2.6% |
| Ecommerce development account not named | 337 of 428 78.7% | 61 of 428 14.3% | 22 of 428 5.1% | 9 of 428 2.1% | 37 of 428 8.6% | 23 of 428 5.4% | 7 of 428 1.6% |
| B2B outbound and lead generation Danish Lead Co | 660 of 752 87.8% | 92 of 752 12.2% | 12 of 752 1.6% | 5 of 752 0.7% | 94 of 752 12.5% | 43 of 752 5.7% | 12 of 752 1.6% |
| Work management software consulting account not named | 387 of 539 71.8% | 63 of 539 11.7% | 5 of 539 0.9% | 4 of 539 0.7% | 97 of 539 18.0% | 28 of 539 5.2% | 10 of 539 1.9% |
| Healthcare benefits, employer buyers account not named | 225 of 255 88.2% | 27 of 255 10.6% | 51 of 255 20.0% | 69 of 255 27.1% | 20 of 255 7.8% | 6 of 255 2.4% | 18 of 255 7.1% |
| Ecommerce analytics software account not named | 541 of 644 84.0% | 55 of 644 8.5% | 83 of 644 12.9% | 1 of 644 0.2% | 127 of 644 19.7% | 35 of 644 5.4% | 30 of 644 4.7% |
| Answer engine optimization SourceRank AI | 461 of 514 89.7% | 23 of 514 4.5% | 44 of 514 8.6% | 12 of 514 2.3% | 102 of 514 19.8% | 52 of 514 10.1% | 39 of 514 7.6% |
| Healthcare benefits, clinic buyers account not named | 211 of 250 84.4% | 11 of 250 4.4% | 39 of 250 15.6% | 37 of 250 14.8% | 27 of 250 10.8% | 10 of 250 4.0% | 5 of 250 2.0% |
Two of the eleven accounts sit in the same broad healthcare category with different buyers, an employer-side set of questions and a clinic-side set, which is why two rows look related. They are separate accounts with separate question sets and separate runs.
Read down the directories column: 54.9%, 45.8%, 16.5%, 14.6%, 14.3%, 12.2%, 11.7%, 10.6%, 8.5%, 4.5%, 4.4%. Read down the documentation column and the order is broadly the reverse. There is no single AI citation profile to optimise against. There are at least eleven.
Vendor sites lead everywhere, so what comes second is the whole story
Vendor and agency websites were the most cited kind in all eleven industries. The column that matters is the next one.
| Industry | Vendor or agency sites | Leading non-vendor kind | Its share | Runner up |
|---|---|---|---|---|
| Coworking and flexible office space account not named | 501 of 641, 78.2% | Directories and marketplaces | 352 of 641, 54.9% | Government and regulator, 29 of 641, 4.5% |
| Local restaurant account not named | 317 of 500, 63.4% | Directories and marketplaces | 229 of 500, 45.8% | Social and video, 29 of 500, 5.8% |
| Healthcare benefits, employer buyers account not named | 225 of 255, 88.2% | Government and regulator | 69 of 255, 27.1% | Docs and standards, 51 of 255, 20.0% |
| Answer engine optimization SourceRank AI | 461 of 514, 89.7% | Social and video | 102 of 514, 19.8% | Community, 52 of 514, 10.1% |
| Ecommerce analytics software account not named | 541 of 644, 84.0% | Social and video | 127 of 644, 19.7% | Docs and standards, 83 of 644, 12.9% |
| Work management software consulting account not named | 387 of 539, 71.8% | Social and video | 97 of 539, 18.0% | Directories and marketplaces, 63 of 539, 11.7% |
| Construction and industrial staffing account not named | 370 of 449, 82.4% | Directories and marketplaces | 74 of 449, 16.5% | Government and regulator, 65 of 449, 14.5% |
| Healthcare benefits, clinic buyers account not named | 211 of 250, 84.4% | Docs and standards | 39 of 250, 15.6% | Government and regulator, 37 of 250, 14.8% |
| Private equity and M&A deal sourcing DealSource Systems | 602 of 687, 87.6% | Directories and marketplaces | 100 of 687, 14.6% | Social and video, 93 of 687, 13.5% |
| Ecommerce development account not named | 337 of 428, 78.7% | Directories and marketplaces | 61 of 428, 14.3% | Social and video, 37 of 428, 8.6% |
| B2B outbound and lead generation Danish Lead Co | 660 of 752, 87.8% | Social and video | 94 of 752, 12.5% | Directories and marketplaces, 92 of 752, 12.2% |
A pattern falls out of that column. In a local, physical, bookable category, coworking and the restaurant, the engine reaches for a marketplace or a listings guide, in roughly half of all answers. In the regulated categories, healthcare benefits on both the employer and the clinic side, it reaches for a government or regulator site and for the professional bodies that set the rules. In industrial staffing it reaches for job boards and the regulator almost equally. In software and B2B services, answer engine optimization, lead generation, ecommerce analytics and work management consulting, it reaches for social and video, which in practice means LinkedIn and YouTube. And in deal sourcing and ecommerce development, where buyers already shop on platforms such as Axial, Clutch and Upwork, directories and marketplaces come first and social second. Same engines, five different answers to the question every buyer of this service asks first.
How far each source kind moves between industries
For each kind, the industry where it was cited most, the industry where it was cited least, and the ratio between them.
| Source kind | All 11 industries | Highest industry | Lowest industry | Gap |
|---|---|---|---|---|
| Government and regulator | 264 of 5,659, 4.7% | Healthcare benefits, employer buyers 69 of 255, 27.1% | Ecommerce analytics software 1 of 644, 0.2% | 174 times |
| Community | 237 of 5,659, 4.2% | Answer engine optimization 52 of 514, 10.1% | Coworking and flexible office space 4 of 641, 0.6% | 16 times |
| Directories and marketplaces | 1,087 of 5,659, 19.2% | Coworking and flexible office space 352 of 641, 54.9% | Healthcare benefits, clinic buyers 11 of 250, 4.4% | 12 times |
| Social and video | 672 of 5,659, 11.9% | Answer engine optimization 102 of 514, 19.8% | Coworking and flexible office space 11 of 641, 1.7% | 12 times |
| Vendor or agency site | 4,612 of 5,659, 81.5% | Answer engine optimization 461 of 514, 89.7% | Local restaurant 317 of 500, 63.4% | 1.4 times |
| Docs and standards | 303 of 5,659, 5.4% | Healthcare benefits, employer buyers 51 of 255, 20.0% | Coworking and flexible office space 0 of 641, 0.0% | no floor: zero answers in the lowest industry |
| News and media | 149 of 5,659, 2.6% | Answer engine optimization 39 of 514, 7.6% | Local restaurant 0 of 500, 0.0% | no floor: zero answers in the lowest industry |
Docs and standards and news and media have no ratio because the lowest industry recorded zero: 0 of 641 coworking answers cited documentation, and 0 of 500 restaurant answers cited a news source.
The aggregate column is the number a vendor would publish as the industry benchmark. Directories and marketplaces were cited in 1,087 of 5,659 answers, 19.2%. That number is true and almost useless: it is nearly three times too low for the coworking operator and more than four times too high for the clinic-side healthcare benefits company. Any single AI citation statistic quoted without an industry attached is describing a company that does not exist.
On social and community sources, the engine matters more than the industry
All eleven industries pooled, split by engine. Share of that engine's answers citing each kind.
| Engine | Answers | No source list | Vendor or agency site | Directories and marketplaces | Docs and standards | Government and regulator | Social and video | Community | News and media |
|---|---|---|---|---|---|---|---|---|---|
| Microsoft Copilot | 1,158 | 0, 0.0% | 1,158 of 1,158 100.0% | 418 of 1,158 36.1% | 76 of 1,158 6.6% | 59 of 1,158 5.1% | 111 of 1,158 9.6% | 15 of 1,158 1.3% | 62 of 1,158 5.4% |
| Google AI Overviews | 1,004 | 50, 5.0% | 936 of 1,004 93.2% | 237 of 1,004 23.6% | 50 of 1,004 5.0% | 37 of 1,004 3.7% | 431 of 1,004 42.9% | 130 of 1,004 12.9% | 42 of 1,004 4.2% |
| Google AI Mode | 1,169 | 131, 11.2% | 881 of 1,169 75.4% | 104 of 1,169 8.9% | 25 of 1,169 2.1% | 27 of 1,169 2.3% | 110 of 1,169 9.4% | 65 of 1,169 5.6% | 14 of 1,169 1.2% |
| ChatGPT | 1,159 | 129, 11.1% | 929 of 1,159 80.2% | 252 of 1,159 21.7% | 134 of 1,159 11.6% | 126 of 1,159 10.9% | 5 of 1,159 0.4% | 5 of 1,159 0.4% | 18 of 1,159 1.6% |
| Gemini | 1,169 | 448, 38.3% | 708 of 1,169 60.6% | 76 of 1,169 6.5% | 18 of 1,169 1.5% | 15 of 1,169 1.3% | 15 of 1,169 1.3% | 22 of 1,169 1.9% | 13 of 1,169 1.1% |
Engine rows use that engine's own answer count as the denominator, because the engines did not all answer the same number of questions.
Three findings sit in that table. First, ChatGPT cited a community site in 5 of its 1,159 answers, 0.4%, and a social or video site in 5 of 1,159, 0.4%. Google AI Overviews cited community in 130 of 1,004, 12.9%, and social or video in 431 of 1,004, 42.9%. On this dataset those are not the same medium. Second, ChatGPT is the engine that reaches for authority: it cited a government or regulator source in 126 of 1,159 answers, 10.9%, against 59 of 1,158 for Microsoft Copilot, 5.1%. Third, Microsoft Copilot cited a vendor or agency site in every one of its 1,158 answers and returned a source list on every single answer.
The ChatGPT community and social figures deserve a caveat rather than a headline. They are a real property of what our collector captured from ChatGPT on these dates, they are consistent across eleven independent accounts, and they match the 0 community citations in 923 ChatGPT answers we recorded in September, which makes a per-account fluke unlikely. It is still one collector and one window of dates, and it may reflect what ChatGPT exposes as a citation rather than what it read.
The same disagreement inside one industry
Two industries where one non-vendor source kind leads, split by engine, to show the effect is not an artefact of pooling.
| Engine | Coworking: answers citing a directory or marketplace | Healthcare benefits, employer buyers: answers citing a government or regulator source |
|---|---|---|
| Microsoft Copilot | 119 of 134, 88.8% | 18 of 51, 35.3% |
| Google AI Overviews | 63 of 86, 73.3% | 13 of 51, 25.5% |
| ChatGPT | 84 of 137, 61.3% | 27 of 51, 52.9% |
| Google AI Mode | 46 of 142, 32.4% | 8 of 51, 15.7% |
| Gemini | 40 of 142, 28.2% | 3 of 51, 5.9% |
In coworking, Microsoft Copilot cited a directory or marketplace in 119 of its 134 answers, 88.8%, while Gemini did so in 40 of 142, 28.2%. In employer-side healthcare benefits, ChatGPT cited a government or regulator source in 27 of 51 answers, 52.9%, and Gemini in 3 of 51, 5.9%. The engine that leads changes with the source kind: Copilot on listings, ChatGPT on regulators. Part of Gemini's low share is that it attached no source list to 73 of its 142 coworking answers, a reminder that how often an engine cites a kind and how often it shows sources at all are two different measurements.
Every industry's full denominator
Including the answers that returned text and no source list, which is the number most AI visibility statistics leave out.
| Industry | Questions | Answers | No source list | Cited links | Distinct domains |
|---|---|---|---|---|---|
| Work management software consulting account not named | 108 | 539 | 146 of 539, 27.1% | 2,422 | 726 |
| Local restaurant account not named | 100 | 500 | 89 of 500, 17.8% | 2,011 | 178 |
| Ecommerce development account not named | 107 | 428 | 67 of 428, 15.7% | 1,987 | 1,254 |
| Ecommerce analytics software account not named | 129 | 644 | 90 of 644, 14.0% | 3,394 | 1,393 |
| Coworking and flexible office space account not named | 142 | 641 | 89 of 641, 13.9% | 3,166 | 696 |
| Private equity and M&A deal sourcing DealSource Systems | 138 | 687 | 79 of 687, 11.5% | 3,580 | 1,471 |
| Construction and industrial staffing account not named | 90 | 449 | 50 of 449, 11.1% | 2,390 | 901 |
| B2B outbound and lead generation Danish Lead Co | 151 | 752 | 81 of 752, 10.8% | 3,970 | 1,436 |
| Healthcare benefits, clinic buyers account not named | 50 | 250 | 25 of 250, 10.0% | 1,262 | 768 |
| Answer engine optimization SourceRank AI | 103 | 514 | 30 of 514, 5.8% | 2,955 | 1,364 |
| Healthcare benefits, employer buyers account not named | 51 | 255 | 12 of 255, 4.7% | 1,461 | 744 |
758 of 5,659 answers, 13.4%, carried no source list at all, and 448 of those came from Gemini. The sourceless rate itself varies by industry, from 12 of 255 answers, 4.7%, up to 146 of 539, 27.1%. Every share on this page is computed over the answers that did carry sources and expressed against the full answer count, so the shares are floors, not estimates. A provider who drops sourceless answers from the denominator will report the same runs with higher numbers.
What to invest in, by industry
What this run says about where citation effort pays and where it does not. Every figure is share of that industry's answers.
| If you are in | Worth the money | Lowest return in this run |
|---|---|---|
| Coworking, flexible space, any bookable local category | Marketplace and directory listings. Cited in 352 of 641 answers, 54.9%, on 59 distinct platforms. The most cited were flexoffice.swiss in 83 answers, workin.space in 62, matchoffice.ch in 46, and immoscout24.ch and instantoffices.com in 44 each | Documentation, 0 of 641 answers. News, 1 of 641. Community, 4 of 641, 0.6% |
| Restaurants and other local venues | Local guides, review sites and delivery platforms. Cited in 229 of 500 answers, 45.8%, on 37 distinct platforms, led by local food and city guides and by Tripadvisor: tripadvisor.dk in 74 answers and tripadvisor.com in 61 | News, 0 of 500 answers. Documentation, 2 of 500, 0.4%. Community, 10 of 500, 2.0% |
| Healthcare benefits, insurance, anything regulated | The government and regulator sources the engines quote, and the professional bodies. On the employer side, government sites were cited in 69 of 255 answers, 27.1%, from 44 government domains led by healthcare.gov, and professional bodies such as naic.org and shrm.org took documentation and standards to 51 of 255, 20.0%. On the clinic side documentation and standards led at 39 of 250, 15.6%, through ama-assn.org and ada.org, with government at 37 of 250, 14.8% | Community, 6 of 255 answers, 2.4% on the employer side. Directories, 11 of 250, 4.4% on the clinic side, all of them comparison and booking sites: eHealth, GoodRx and Zocdoc |
| Staffing and industrial services | Job boards plus the regulator. Directories and job boards cited in 74 of 449 answers, 16.5%, led by ziprecruiter.com in 20 and clearlyrated.com in 14, and government in 65 of 449, 14.5%, led by dol.gov in 19 and osha.gov in 18 | Community, 4 of 449 answers, 0.9%. News, 9 of 449, 2.0% |
| Software and B2B services: AEO, lead generation, ecommerce analytics, work management consulting | Social and video, which here means YouTube and LinkedIn. Cited in 12.5% to 19.8% of answers across these four accounts. Review directories earn a place in two of them: in lead generation clutch.co was cited in 44 of 752 answers, and in ecommerce analytics g2.com in 23 of 644 and gartner.com in 21 | Government, under 3% in all four. For answer engine optimization specifically, directories were 23 of 514 answers, 4.5% |
| Deal sourcing and ecommerce development | Marketplaces first, social second. Directories and marketplaces cited in 100 of 687 deal sourcing answers, 14.6%, led by axial.net in 22, and in 61 of 428 ecommerce development answers, 14.3%, led by clutch.co and upwork.com in 16 each. Social and video followed at 13.5% and 8.6% | Government, 20 of 687 and 9 of 428 answers, under 3% in both. News, 2.6% and 1.6% |
The general rule this run supports: your own website carries the citation load in every industry, between 63.4% and 89.7% of answers, and no listing, profile or press mention substitutes for it. What changes by industry is the second surface, and that is where a budget is either well spent or wasted. A coworking operator ignoring marketplaces is invisible in 54.9% of the answers about it. An answer engine optimization company buying directory listings is buying into 4.5%.
What this does not tell you
- +One industry, one company, one run. Each industry here is represented by a single company's tracked question set on a single date. A second coworking operator, or the same one a week later, would not return identical shares. Treat each row as one observation, not as the industry's constant.
- +The question sets are not identical. Each account has its own questions, from 50 to 151 of them, written for its own buyers. Part of any gap between two industries is a difference in what was asked, not only in what the engines did with it.
- +Domain kind is a judgement. Seven kinds, assigned by an explicit named rule set printed in the SQL at the end of this page's source, with vendor or agency site as the residual. Local food guides and tourism portals count as directories, because their purpose is to list third parties. The hardest calls are the ones where a category's competitors are the source type. We classified by what a site primarily is and published the list so it can be re-cut.
- +The vendor bucket includes the company itself. Vendor or agency site covers each measured company's own website, its competitors and unrelated commercial sites. We do not publish any individual account's own-domain figures, so the vendor column should be read as a category total, not as anyone's visibility score.
- +Sourceless answers are real and uneven. 758 of 5,659 answers, 13.4%, carried no source list, 448 of them from Gemini, and the rate ranges from 4.7% to 27.1% between industries. Shares here are floors.
- +It measures sources, not sales. Being cited is not being chosen, and a marketplace citation is not a booking. This page counts what the engine reached for.
- +Eleven industries is not every industry. There is nothing here about legal services, industrial manufacturing or consumer retail. The finding that the mix varies is well supported. The specific number for an industry not listed is not something this page can give you.
We re-measure every account every 2 days. This page is refreshed in place on the same URL, so the URL stays stable and the numbers stay current rather than freezing on the day we first published them. If a figure here is being quoted, quote it with the run dates attached.
Key takeaways
- +Directories and marketplaces were cited in 352 of 641 answers, 54.9%, for a coworking operator and 11 of 250 answers, 4.4%, for a healthcare benefits company selling to clinics. Same method, same five engines, same week.
- +Government and regulator sources ran from 69 of 255 answers, 27.1%, in employer-side healthcare benefits to 1 of 644, 0.2%, in ecommerce analytics software. That is the widest measurable gap on the page, 174 times.
- +Vendor and agency websites were the most cited kind in all eleven industries and moved least: 317 of 500 answers, 63.4%, to 461 of 514, 89.7%.
- +ChatGPT cited a community site in 5 of its 1,159 answers. Google AI Overviews cited one in 130 of its 1,004, 12.9%. Microsoft Copilot cited a vendor site in all 1,158 of its answers.
- +758 of 5,659 answers, 13.4%, carried no source list, which is why the denominator has to be published alongside any AI citation statistic.
Cite this research
Journalists, analysts and writers are welcome to quote, chart and republish the figures on this page. Please credit SourceRank AI and link to this page as the source. For the underlying data, a comment from Frederik Jakobsen or a question we have not covered, see the press page.
Suggested citation
SourceRank AI (2026). What AI engines cite changes completely by industry. Measured 21 to 23 October 2026. https://sourcerankai.com/research/what-ai-engines-cite-by-industry/
Link to copy
<a href="https://sourcerankai.com/research/what-ai-engines-cite-by-industry/">What AI engines cite changes completely by industry (SourceRank AI, 2026)</a>Where to go next
What AI engines actually cite
The companion report. One of these eleven runs taken apart domain by domain, including the 25 most cited domains and the review directory count.
The measurement method
How the tracked question sets, the engines and the snapshots work, published before any numbers came out of them.
Who AI recommends
Which firms ChatGPT, Gemini and Google AI name in 22 B2B categories, from a dated October 2026 run.
To see how the engines answer your own buyers' questions rather than one of these eleven industries, start with the free AI visibility audit. The free site readiness check is a scan of your site, not a citation measurement, so it will not produce numbers like the ones on this page.
Common questions
Does the mix of sources AI engines cite change by industry?
Yes, and by far more than the averages suggest. Across eleven companies in eleven industries, 5,659 answers on five engines, directories and marketplaces were cited in 352 of 641 answers, 54.9%, for a coworking operator, and in 11 of 250 answers, 4.4%, for a healthcare benefits company selling to clinics. Government and regulator sources ran from 69 of 255 answers, 27.1%, in employer-side healthcare benefits down to 1 of 644, 0.2%, in ecommerce analytics software.
What do AI engines cite in every industry?
Vendor and agency websites. They were the most cited kind in all eleven industries, and the range was narrow: from 317 of 500 answers, 63.4%, at the low end to 461 of 514 answers, 89.7%, at the high end. That is a spread of 1.4 times, against 174 times for government and regulator sources.
Should a company invest in directory listings to get cited by AI?
It depends entirely on the industry, which is the point of this page. For the coworking operator, directories and marketplaces were cited in 352 of 641 answers, 54.9%, and for a local restaurant in 229 of 500, 45.8%, so listings are a primary surface. For the clinic-side healthcare benefits company they were cited in 11 of 250 answers, 4.4%, and for the answer engine optimization company in 23 of 514, 4.5%, so the same spend buys close to nothing.
Do different AI engines cite different kinds of sources?
Sharply. ChatGPT cited a community site such as Reddit, Quora or Stack Overflow in 5 of its 1,159 answers, 0.4%, and a social or video site in 5 of 1,159, 0.4%. Google AI Overviews cited community in 130 of its 1,004 answers, 12.9%, and social or video in 431 of 1,004, 42.9%. Microsoft Copilot cited a vendor or agency site in all 1,158 of its answers.
Which AI engines does SourceRank AI measure?
Five: ChatGPT, Gemini, Microsoft Copilot, Google AI Mode and Google AI Overviews. Every number on this page comes from those five engines and no others.