# Evidence: Market size 16 of 108 entries in the collection “Evidence” by Robert Haase, as of 21 September 2026. Page: https://robert-haase.de/en/evidence.html · Overview of all claims: https://robert-haase.de/en/evidence.md · JSON: https://robert-haase.de/en/evidence.json · Deutsch: https://robert-haase.de/belege-markt.md License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Please cite the primary source, not this page. This file is generated from the page. Where the two differ, the page applies. ## How this collection is built Every figure is traced back to the body that measured it, not to the article citing it. On their way through the retellings, figures lose their denominator first, then their caveat, and finally their origin. Where a figure is only accessible through a third party, that intermediary is named in the source line. Own measurements carry their method with them; they have not been independently verified yet. **The limit belongs to the number.** The most common error is not the wrong number but the right one carrying a claim that reaches further than the evidence. That is why every entry has two parts, and the second one matters more. Above each figure sits what kind of evidence it is, from verified study to single case. That decides how far it carries. What does not survive the check does not get in, or gets taken out, my own articles included. One of them claimed that 44 percent of US online shoppers begin their purchase journey in a language model, attributed to Bain. Bain gives two other figures, 17 percent and 30 to 45 percent, which had merged into one along the way. Both are here now; the 44 is not. This page ages. Every entry carries its date; superseded numbers get replaced, not quietly deleted. If you find an error, [write to me](mailto:hallo@robert-haase.de) and I will correct it and note the date. The collection does not map the state of the research, only the figures I needed for my own texts. Free to use with attribution. When in doubt, link the primary source rather than this page. ## Grades in this topic + Verified study, vendor documentation, or court decision (2) → ki-nutzung-deutschland, ki-anteil-artikel + Preliminary: prototype, single test, forecast, or vendor figure (8) → machine-customers, marktgroesse, mcp-verbreitung, agentenhandel-2030, dark-data-55, ki-verkehrsanteil, markenklone, suchmarkt-wachstum + Status, case report, or market observation (6) → nicht-menschlicher-verkehr, cmo-ki-anteil, geo-verbreitung, in-house-verlagerung, insourcing-absicht, agentur-selbstbild --- ## machine-customers **Claim:** In a Gartner survey, chief executives estimate that by 2030, 15 to 20 percent of their revenue will come from machine customers. **What this is:** a self-assessment by surveyed executives about the future, not a measurement and not a Gartner house forecast. Citations routinely compress both into "Gartner expects". Estimates like this for new categories are often wrong, usually in the timing rather than the direction. **On sourcing:** the page blocks automated retrieval; the wording is evidenced through an archive and Gartner’s own video title, not through a direct fetch. **Source:** Gartner, CEO survey, cited in the Think Again series · archive snapshot July 2026 · [Source](https://www.gartner.com/en/experts/think-again-series/machine-customers) **Grade:** Self-assessment, forecast · Group: Preliminary: prototype, single test, forecast, or vendor figure **Permalink:** https://robert-haase.de/en/evidence.html#machine-customers --- ## marktgroesse **Claim:** The large user numbers for AI systems come from the vendors themselves and are not comparable with each other: around 900 million weekly active users for ChatGPT, over one billion monthly users for Google AI Mode. **Why the comparison limps:** One number counts **weekly**, the other **monthly**. Placed side by side they still read as equivalent, and that is exactly how they travel through presentations. Neither is independently audited. The Google figure is at least documented by the vendor directly; the ChatGPT figure circulates as a company statement in reports about it. Usable as an order of magnitude, not as evidence. **Source:** Google, AI Mode blog post, 19 May 2026 · ChatGPT figure: OpenAI statement, February 2026 · [Source](https://blog.google/products-and-platforms/products/search/search-io-2026/) **Grade:** Vendor figures · Group: Preliminary: prototype, single test, forecast, or vendor figure **Permalink:** https://robert-haase.de/en/evidence.html#marktgroesse --- ## ki-nutzung-deutschland **Claim:** 26 percent of German companies with ten or more employees used AI technologies in 2025. Among large companies with 250 or more employees it is 57 percent, among small ones 23 percent. **What the number does not say:** It is collected as a yes-or-no and says nothing about intensity, maturity, or effect, and it is not broken down by function — so it is no evidence about marketing or brand management. **Why it belongs here anyway:** it is the most methodologically rigorous German figure available and a sober anchor against industry-association surveys reporting markedly higher numbers for the same year. When two figures on the same question diverge widely, the cause lies in population and method, not in reality. **Source:** German Federal Statistical Office, ICT usage survey, companies with 10 or more employees · November 2025 · [Source](https://www.destatis.de/DE/Themen/Branchen-Unternehmen/Unternehmen/IKT-in-Unternehmen-IKT-Branche/Tabellen/ikti-unternehmen-kuenstliche-intelligenz.html) **Grade:** Official statistics · Group: Verified study, vendor documentation, or court decision **Permalink:** https://robert-haase.de/en/evidence.html#ki-nutzung-deutschland --- ## nicht-menschlicher-verkehr **Claim:** Cloudflare reports that in 2026, for the first time, more than half of Internet traffic is not human. Better quantified in the same report: 52 percent of crawler requests served AI model training in June 2026, up from 22 percent in spring 2025. **What the number does not say:** For the majority claim Cloudflare states *neither what traffic was measured* — page requests, all requests? — *nor over what period*. It rests on their own network, which is large but is not the Internet. The crawler figure is dated and carries a prior-year comparison, making it the more usable of the two. **Care when reusing:** secondary sources circulate the figure as “57.5 percent” — that number appears nowhere at Cloudflare. **Source:** Cloudflare, “Content Independence Day, one year on”, 1 July 2026 · data basis per the report: Cloudflare Radar and Investor Day 2026 · [Source](https://blog.cloudflare.com/agentic-internet-bot-report/) **Grade:** Market observation · Group: Status, case report, or market observation **Permalink:** https://robert-haase.de/en/evidence.html#nicht-menschlicher-verkehr --- ## mcp-verbreitung **Claim:** On handing the Model Context Protocol to the Linux Foundation on 9 December 2025, Anthropic gives more than 10,000 active public MCP servers and over 97 million monthly SDK downloads across Python and TypeScript. The platinum members of the new Agentic AI Foundation include, per the foundation, AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI. **What the number does not say:** it sits in the developer’s own donation post, in a list under the heading “incredible adoption”, with no counting rule at all: no registry, no as-of date, no definition of “active”. Cross-checking helps only so far, since no authoritative registry exists: a complete dump of the official registry on 10 September 2026 gives 30,363 registered servers, 30,031 of them in state “active”. It cannot be set against the vendor figure: Anthropic gives no counting rule and means a different object, so no growth rate follows. And the registry counts entries someone created, not servers in use; it is marked a preview and states itself that one should assume “minimal-to-no moderation”. And a server is not a user. **Care when passing it on:** the Linux Foundation calls the same figure “published”, turning active servers into published ones. **On the membership list:** platinum membership is paid: that AWS, Google, Microsoft and OpenAI carry the governance does not establish that they use the protocol in their products. The foundation writes “include”, so the list is not exhaustive. State of play: December 2025. **Source:** Anthropic, “Donating the Model Context Protocol and establishing the Agentic AI Foundation”, 9 December 2025, the originating source for both figures · Linux Foundation, press release on the founding of the Agentic AI Foundation, 9 December 2025, for the membership list ([press release](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation)) · own cross-check against the official registry registry.modelcontextprotocol.io on 10 September 2026 · [Source](https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation) **Grade:** Vendor figures · Group: Preliminary: prototype, single test, forecast, or vendor figure **Permalink:** https://robert-haase.de/en/evidence.html#mcp-verbreitung --- ## agentenhandel-2030 **Claim:** Morgan Stanley puts agentic shoppers at 190 to 385 billion dollars of US e-commerce by 2030, a likely 10 percent market share and up to 20 percent in the optimistic case. Nine days later Bain puts agentic commerce at 300 to 500 billion dollars, roughly 15 to 25 percent. Only Bain states an inclusion rule and counts influenced purchases. **Why the overlap is not agreement:** Bain counts purchases “initiated, influenced, or completed” by agents and excludes only journeys using nothing but AI-assisted search or discovery. Morgan Stanley states no rule; its assistants search, compare prices and anticipate repeat purchases, with minimal user intervention. Recalculated: both imply roughly two trillion dollars of online retail, so the difference sits largely in the numerator. **What no figure says:** how much the agent closes itself. Bain’s closing line puts AI at “up to a quarter of transactions”, the same upper bound as the share figure, counted in transactions rather than sales. Both sell advice on this, neither gives a base or a method for 2030, only adoption figures are sourced, both figures cover the US only. **Source:** Morgan Stanley Research, “Here Come the Shopping Bots”, house forecast for US online retail · 8 December 2025 · Bain & Company, Snap Chart “2030 Forecast: How Agentic AI Will Reshape US Retail” by Aaron Cheris, Mikey Vu, Stephanie Koszyk and Katherine Hall, house forecast for US online retail · 17 December 2025 · bain.com blocks automated retrieval, HTTP 403 on 10 September 2026, read in the archive capture of 17 December 2025 · [Archive capture](https://web.archive.org/web/20251217213656/https://www.bain.com/insights/2030-forecast-how-agentic-ai-will-reshape-us-retail-snap-chart/) · [Source](https://www.morganstanley.com/insights/articles/agentic-commerce-market-impact-outlook) **Grade:** House forecasts, not comparable · Group: Preliminary: prototype, single test, forecast, or vendor figure **Permalink:** https://robert-haase.de/en/evidence.html#agentenhandel-2030 --- ## cmo-ki-anteil **Claim:** Marketing leaders at U.S. companies use AI or machine learning 24.2 percent of the time they spend optimizing and automating marketing. The typical company says 20 percent. Two surveys earlier the figures were 13.1 (September 2024) and 17.2 percent (early 2025). For generative AI alone the figure rose from 7.0 through 15.1 to 22.4 percent. Within three years the same respondents expect 55.9 percent. **What the number does not say:** It measures a self-estimated share of time, averaged across respondents and never checked against system data. It is neither a share of companies nor a share of budget. 24.2 is the mean of a right-skewed distribution; the median is 20. The question was answered by 191 of the 2,111 people invited, about 9 percent, the expectation question by 188. The same question produced 34.5 and then 44.2 percent in the two preceding waves; the expectation climbs with every wave and none has ever been checked. **The sector figures belong to two questions:** The 36.1 percent comes from the overall question and the largest sector cell (40 companies), the 8.7 percent from the generative AI one and one of the smallest (3). For the overall question the report gives no low at all. **Source:** The CMO Survey, 35th edition, conducted by Christine Moorman at Duke University’s Fuqua School of Business, sponsored by Duke, Deloitte and the American Marketing Association · 2,111 marketing leaders at U.S. for-profit companies invited, 308 responses, 14.6 percent response rate, 191 of them to this question, 97 percent VP level or above · fielded 7 to 29 January 2026, report published April 2026 · Highlights Report page 21 (AI and machine learning) and page 23 (generative AI), figures in the Topline Report page 12, sector cell sizes in the Firm and Industry Breakout Report pages 36 and 45 · [Breakout Report](https://cmosurvey.org/wp-content/uploads/2026/03/The_CMO_Survey-Firm_and_Industry_Breakout_Report-2026-1.pdf) · [Source](https://cmosurvey.org/wp-content/uploads/2026/04/The_CMO_Survey-Highlights_and_Insights_Report-2026.pdf) **Grade:** Survey · Group: Status, case report, or market observation **Permalink:** https://robert-haase.de/en/evidence.html#cmo-ki-anteil --- ## dark-data-55 **Claim:** The most quoted figure on unused corporate data, 55 percent, bundles self-estimates the 1,357 respondents made about their own organisation. It was fielded in 2018/19 by the research arm of the PR agency FleishmanHillard on behalf of Splunk, a vendor selling software to analyse exactly this data. **What the number actually covers:** Percent of what stays open, and the report names neither unit nor period nor the statistic used. The only definition put to respondents was “Information that can be captured, quantified and analyzed”. On pages 3 and 12 it reads as a fact, only the regional and country sections from page 10 onwards reveal it as an estimate: US 56 percent, Germany 53, China 50 “compared with a global 55 percent”. An estimate across organisations is not a share of any total, and 1,357 is the sum of the market samples, while the report says 1,300 respondents. **What gets routinely mixed in with it:** 60 percent of respondents say half or more of their data is dark, 33 percent say 75 percent or more. Those are shares of respondents, not shares of data. Splunk still circulates the figure without a year, on 26 March 2025 as “recent” and there without a sample either. **Source:** TRUE Global Intelligence (FleishmanHillard) for Splunk, The State of Dark Data, 1,357 respondents from IT and business across seven countries · fielded October 2018 to January 2019, report May 2019 · [Source](https://www.splunk.com/content/dam/splunk2/en_us/gated/white-paper/the-state-of-dark-data.pdf) **Grade:** Usually miscited · Group: Preliminary: prototype, single test, forecast, or vendor figure **Permalink:** https://robert-haase.de/en/evidence.html#dark-data-55 --- ## geo-verbreitung **Claim:** 78 of 188 US companies that answered this question use AI for generative engine optimization, that is, to get their own content to appear in AI-generated search answers. That is 41.5 percent, with a 95 percent confidence interval of plus or minus 7.1 percentage points. **What the number does not say:** It is self-reported, comes from a check-all-that-apply question and records the doing, not the result. The denominator is the trap: it is not the 308 respondents but the 188 who answered this question; all 188 ticked at least one box (response percent 100.0). No company using no AI at all sits in the denominator, so the 41.5 percent is a share among AI users. Computing against 308 yields 128 instead of 78. With the interval the range runs from 34 to 49 percent: a good four in ten, not one in two. **Who was asked:** US companies only, 97 percent at VP level or above, 308 of 2,111 people contacted, a 14.6 percent response rate. The figure does not transfer to the German market. What is new is the answer option, not the question: GEO was on the list for the first time in 2026 and has no comparison value, while the question itself is reported as a time series against Fall 2023. **Source:** The CMO Survey, 35th edition, run at the Fuqua School of Business, Duke University, sponsored by Duke, Deloitte and the American Marketing Association · Topline Report 2026, page 12, answer option “GEO (i.e., Generative Engine Optimization to get content to appear in AI-generated search results)”: 78 of 188 cases, 41.5 percent, plus or minus 7.1 percentage points · 308 respondents out of 2,111 marketing leaders contacted at US for-profit companies · fielded 7 to 29 January 2026 · take care when looking it up: the row above, “Predictive analytics for customer insights”, carries the same 41.5 percent and likewise 78 cases · [Source](https://cmosurvey.org/wp-content/uploads/2026/03/The_CMO_Survey-Topline_Report-2026.pdf) **Grade:** Survey · Group: Status, case report, or market observation **Permalink:** https://robert-haase.de/en/evidence.html#geo-verbreitung --- ## in-house-verlagerung **Claim:** In the member survey run by the US advertising association ANA, 82 percent of the members surveyed said in 2023 that they had an in-house agency, after 78 percent in 2018, 58 percent in 2013 and 42 percent in 2008. 65 percent said in 2023 that they had moved ongoing business from an external agency in-house in the preceding three years. In 2018 it was 70 percent, in 2013 only 56. **What the figures do not say:** They measure where work sits, not whether it gets better or more on-brand. An industry association surveys its own members, participation is voluntary, and respondents may belong to the in-house agency themselves. **Careful with the 65 percent:** The ANA states a base for each question, regularly smaller than the participant count. In 2018 the base for this question was 166 of 412 respondents; for 2023 it is not published and must not be applied to the 162 participants. The gap between 70 and 65 percent carries no turning point: the interval runs from 63 to 77 percent in 2018 and, on at most 162 answers, from 58 to 72 percent in 2023, and the waves differ in size. The ANA report of June 2026 does not continue the series, it surveys award jurors; the next member wave would be 2028. **Source:** Association of National Advertisers, member survey “The Continued Rise of the In-House Agency: 2023 Edition”, 162 respondents, fielded February and March 2023 · press release of 2 May 2023, the report itself sits behind the membership wall · comparison values and per-question bases from the predecessor report of October 2018, 412 respondents, 57 pages · [Predecessor report](https://www.ana.net/content/show/id/pr-2018-inhouse-rising) · [Source](https://www.ana.net/content/show/id/79185) **Grade:** Survey · Group: Status, case report, or market observation **Permalink:** https://robert-haase.de/en/evidence.html#in-house-verlagerung --- ## ki-anteil-artikel **Claim:** Of the English-language articles newly published in the first quarter of 2026, 49.9 percent were primarily AI-generated. Since early 2025 the share has moved between 44.6 and 50.9 percent, with no upward trend. **What the number does not say:** It measures newly published English-language articles and listicles carrying article markup and at least 100 words, drawn from Common Crawl, not the web as a whole, other languages, social media or video. And nothing about reach: Graphite collected the data in June 2025 and published it in October 2025, finding that 86 percent of articles ranking in Google across 31,493 keywords and 82 percent of those cited by ChatGPT and Perplexity were written by humans, but measured with a fourth detector (Surfer, false positive rate 4.2 percent), so not on the same scale. **How firm the 50 percent is:** the figure averages three detectors that diverge by 6.4 points in the same quarter (Pangram 47.7, Copyleaks 48.1, GPTZero 54.1). The spread is wider than the distance to the 50 percent mark, and two of the three put humans ahead. GPTZero counts a “mixed” verdict entirely on the AI side (6.4 percent of articles); Pangram and Copyleaks go by which share is larger. **Source:** Graphite, growth agency · about 55,400 English-language articles randomly drawn from Common Crawl, carrying article markup and at least 100 words, published between January 2020 and March 2026 · classified by averaging three detectors (Pangram, Copyleaks, GPTZero), whose false positive rates of 1.844, 1.836 and 1.355 percent were measured on about 15,700 articles from the same sample published before ChatGPT · quarterly data openly available · May 2026 · [Source](https://graphite.io/five-percent/research/ai-now-writes-as-many-online-articles-as-humans-do) **Grade:** Verified study · Group: Verified study, vendor documentation, or court decision **Permalink:** https://robert-haase.de/en/evidence.html#ki-anteil-artikel --- ## ki-verkehrsanteil **Claim:** Three analytics vendors put a number on the share of website visits that arrive from an AI assistant: Contentsquare 0.2 percent in the fourth quarter of 2025, Semrush 0.14 percent for the year 2025, Conductor 1.08 percent for May to September 2025. Three separately collected measurements, fractions of a percent up to a good one percent. **What the numbers do not say:** They count clicks arriving with an identifiable AI referral source, not how often a brand is named in answers. Google AI Mode sits outside the two values that address it: Semrush tracks it as a separate channel at 0.01 percent, and Conductor notes that Google Analytics does not separate it from organic traffic. **Why the three values do not belong side by side:** Contentsquare measures 6,500 websites worldwide, Semrush more than 50,000 worldwide, Conductor 1,215 of its own customer domains in the US and calls its figures averages. The near eightfold gap between them, 0.14 to 1.08 percent, is largely a question of who was measured. All three sell analytics tools; Contentsquare and Conductor measure their own customer base, Semrush estimates from a bought-in clickstream panel. None of the values is independently audited. **Source:** Contentsquare, 2026 Digital Experience Benchmarks, 99 billion sessions and 6,500 websites worldwide, fourth quarter 2024 against fourth quarter 2025 · 29 January 2026 · Semrush, Traffic & Market Toolkit, more than 50,000 websites and 17 industries worldwide, January to December 2025 · 27 April 2026 · Conductor, AEO/GEO Benchmarks, traffic section from 1,215 of its own enterprise customer domains in the US, May to September 2025 · last updated 6 July 2026 · [Source](https://www.semrush.com/blog/traffic-channel-mix-study/) **Grade:** Three vendor measurements · Group: Preliminary: prototype, single test, forecast, or vendor figure **Permalink:** https://robert-haase.de/en/evidence.html#ki-verkehrsanteil --- ## markenklone **Claim:** Takedown provider Netcraft states that between March 2024 and March 2025 it acted against 1.3 million phishing sites imitating more than 16,000 organisations. **What the number does not say:** It does not say these sites are gone. Netcraft writes “disrupted” and folds blocking and removal into that one word, how the 1.3 million splits is stated nowhere. That same page tells readers to ask vendors exactly this. It also says nothing about whether a published brand specification makes cloning easier or harder, for which there is no comparison group. All it establishes is that clones exist in bulk. **Who counted:** Netcraft itself, its own operations, on a page selling that service. Nothing is independently audited. The world share of about a third often quoted alongside is therefore not in the claim: the page never quantifies the world total and contradicts itself, once a share of takedowns, once of attacks. Nor are the 16,000 organisations a market size. **Source:** Netcraft, guide to detecting and disrupting phishing websites, the vendor’s own operations from March 2024 to March 2025 · 12 December 2025, last modified 12 March 2026 · [Source](https://www.netcraft.com/guide/phishing-website-detection-disruption) **Grade:** Vendor figures · Group: Preliminary: prototype, single test, forecast, or vendor figure **Permalink:** https://robert-haase.de/en/evidence.html#markenklone --- ## suchmarkt-wachstum **Claim:** Between the first quarter of 2023 and the fourth quarter of 2025, search engine visits and search-like AI sessions combined grew by 26 percent worldwide, from 82.0 to 103.2 billion per month. Google’s share falls from 89 to 71 percent, ChatGPT reaches 20 percent. **What the number does not say:** It counts not searches but visits and sessions: a Google visit 6.7 page views on average, an app session an unknown number of prompts. And unevenly: search engines web only, AI web and app. Search apps it excludes as “relatively low”, without a figure, though 83 percent of AI use is in apps. Of AI, only the 52 percent of “asking” count, which the authors call an upper bound. The 71 percent are Google Search and Gemini combined; per day the growth is 23.1, not 26. It cites those same 26 percent elsewhere for 2025 against 2024, where its open raw data yields 17.3. **Where the data comes from:** Similarweb estimates without server measurement, validated only for the search engine figures, across eight websites, only as a trend correlation; the AI figures not against first-party data at all. The author sells visibility in search engines and AI answers and discloses that calling both large serves him. **Source:** Graphite (Ethan Smith) · analysis of Similarweb estimates for web visits and app sessions worldwide, July 2020 to December 2025, raw data public · March 2026 · [Source](https://graphite.io/five-percent/research/ai-is-much-bigger-than-you-think) **Grade:** Market observation on estimated data · Group: Preliminary: prototype, single test, forecast, or vendor figure **Permalink:** https://robert-haase.de/en/evidence.html#suchmarkt-wachstum --- ## insourcing-absicht **Claim:** Asked "Do you plan to cover more marketing services in-house through AI?", 80.0 percent of 170 executives at German companies with budget and decision authority answer yes, 11.2 percent no, and 8.8 percent do not know. Across company sizes the intention is stable: 79 percent at companies with 100 to 999 employees, 81 percent at 1,000 and above. **What the figure does not say:** What is measured is a plan, not a move, and the question names the cause itself; a yes has hired no one. What counts as a marketing service is left to each respondent. 170 self-reports from an online survey, with no information on the population or the response rate; the industry breakdowns in the same study explicitly rest on small numbers. The paper is published by the German agency association GWA, and the survey was run by the Handelsblatt Research Institute. By its own account the study claims no representativeness and permits no firm causal statements. The ANA series on this page measures something else, namely moves already made; there the share has recently fallen from 70 to 65 percent. **Source:** GWA KI-Whitepaper 2026, "KI-Studien" section, question 6 · 170 executives at German companies with budget and decision authority · 1 to 9 April 2026 · descriptive online survey with self-reports · conducted by Handelsblatt Research Institute and techconsult in cooperation with GWA · [Source](https://www.gwa.de/content/uploads/2026/09/GWA-KI-Whitepaper-2026-KI-Studien.pdf) **Grade:** Survey · Group: Status, case report, or market observation **Permalink:** https://robert-haase.de/en/evidence.html#insourcing-absicht --- ## agentur-selbstbild **Claim:** 96.2 percent of 78 executives from member agencies of the German agency association GWA rate their own agency's AI maturity as "advanced" (71.8 percent) or "expert" (24.4 percent), 3.8 percent as "beginner". The same respondents rate the average level of AI knowledge across the agency industry in Germany, Austria and Switzerland mostly at 3 on a scale of 1 to 5 (57.7 percent), 19.2 percent at 2 and 23.1 percent at 4; nobody picks the extremes 1 or 5. **What the figures do not say:** Both are self-assessments, not a test and not a comparison with actual use. The two questions are not comparable: one's own maturity is asked in three steps, beginner, advanced, expert, the industry's level of knowledge on a scale from 1 for very poor to 5 for very good; there is no conversion between them. That the same people rate themselves above their surroundings is plausible, it is not measured. The 3.8 percent are three agencies. 78 answers from within an association, given voluntarily; those who take part are working on the topic. The comparison with the previous wave of 2024/25 does not hold, that sample was smaller and differently composed (n = 52). **Source:** GWA KI-Whitepaper 2026, "KI-Studien" section · 78 executives from GWA member agencies · 12 February to 6 March 2026 · descriptive online survey with self-reports · published by GWA Tech & Innovation Forum · questions: "How would you rate your agency's current level of AI maturity?" and "How do you rate the current average level of AI knowledge across the agency industry in the DACH region?" · [Source](https://www.gwa.de/content/uploads/2026/09/GWA-KI-Whitepaper-2026-KI-Studien.pdf) **Grade:** Survey · Group: Status, case report, or market observation **Permalink:** https://robert-haase.de/en/evidence.html#agentur-selbstbild --- End of file: 16 of 16 entries on Market size. Last entry: agentur-selbstbild.