SISTRIX today published data showing that AI-generated B2B texts score an average of just 44.1 points on a 100-point readability and quality scale, with none of the three tested language models - Claude, Gemini and ChatGPT - reaching what the measurement framework classifies as an acceptable range, while a companion set of tools now lets marketers see how those same AI systems talk about their brand.

The findings arrived in SISTRIX's July 2026 monthly newsletter, sent on July 31, 2026, by Johannes Beus, founder and managing director of the Bonn-based search analytics company. The newsletter, titled "Monatsrückblick Juli: Sentiment Analyse im AI Check, wie gut sind KI Texte?, Meetup in Hamburg + mehr," covered three distinct subjects: a new set of brand-perception analyses inside the company's AI Check product, a joint study with content-marketing firm Wortliga measuring the quality of AI-generated business text, and a summary of product updates SISTRIX shipped during the second quarter of 2026.

AI text quality study finds wide gaps tied to prompting, not models

SISTRIX and Wortliga examined 2,112 AI-generated B2B texts produced with Claude, Gemini and ChatGPT, according to the newsletter. The texts were scored using the Wortliga-Score, a 100-point readability and quality metric, with 60 points marking the threshold for what the study classifies as an acceptable, "green" result.

None of the three models reached that threshold on average. The overall mean across all texts and models sat at 44.1 points. Claude scored highest at 47.7 points and, according to SISTRIX, produced the most consistent results. Gemini followed at 46.8 points, though the newsletter noted it tends toward filler words that inflate length without adding substance. ChatGPT trailed at 37.7 points, a gap of roughly ten points below the two other models.

The study's more consequential finding, however, concerned the effect of prompt design rather than model choice. According to SISTRIX, scores for the same model swung between 1 and 96 points depending entirely on how the prompt was written. A bureaucratically phrased instruction produced text that was itself difficult to read, while a simple instruction such as "write clearly" raised the readability component to 79.4 points - but only by sacrificing depth of subject-matter detail, the newsletter reported.

That trade-off matters for a marketing and content-operations audience for a clear reason: it reframes AI text quality as a solvable instruction-design problem rather than a fixed limitation of any single vendor's model. SISTRIX said further detail on which prompt rules performed best in the test, and how those rules translate across different text categories, appears in a companion blog post on sistrix.de. The newsletter did not specify the exact date the underlying texts were generated, nor the industries or content types the 2,112 documents spanned beyond the general B2B category.

The timing places this research alongside a broader body of evidence PPC Land has tracked regarding AI-generated content quality across the advertising and publishing industries. Integral Ad Science reported in July 2025 that 75 percent of surveyed advertisers did not want their ads appearing next to low-quality AI-generated content, a concern the company later moved to address with a dedicated detection feature that entered open beta in April 2026. Separately, a 2026 analysis from TAG and the ANA found that AI-generated slop content received premium content grades 70 percent of the time under existing classification systems, underscoring how difficult automated quality assessment remains even as the underlying content volume grows.

Three new AI Check analyses target the gap between self-image and AI perception

Alongside the text-quality study, SISTRIX introduced three new analyses inside its AI Check tool, a product that lets users enter a domain or brand name and see how it is described, cited or recommended across AI-generated answers.

According to the newsletter, the first of the three - a Sentiment Analysis - calculates a balance of praise and criticism drawn from actual AI responses, broken down by platform and topic. The example shown for the automotive brand Porsche recorded a sentiment score of positive 46, described as "overwhelmingly positive," with 726 instances of praise against 271 instances of criticism. The platform breakdown in that example showed Google's AI Mode contributing a sentiment score of positive 67, AI Overview at positive 48, and ChatGPT at positive 7 - a considerably narrower margin between praise and criticism on OpenAI's platform than on Google's two AI surfaces.

The second tool, a Competitive Perception analysis, places the attributed strengths and weaknesses of multiple brands side by side, on the reasoning that whether a given trait counts as an advantage becomes clear only in comparison with rivals. The third, a Recommendation Map, goes further still: it records which specific purchase occasions a brand is recommended for as the first choice, mentioned as an option, or simply absent from entirely.

All three analyses, according to SISTRIX, are built from a consistent methodological base: no claim originates from a model's own stated opinion, but from real AI-generated responses to real user questions, each one documented with the original citation and the prompt that produced it. SISTRIX said full details on the new features are available in the company's changelog.

The newsletter also referenced a separate, previously published capability called the Brand-Gap-Analysis, which SISTRIX described as identifying where a brand's self-image and its AI-perceived image diverge, whether in attributed strengths and weaknesses or in purchase situations where a brand does not appear at all. According to the company, this divergence can be used to inform concrete brand-messaging adjustments intended to bring AI-generated descriptions closer to a brand's desired self-image, with a separate tutorial covering implementation.

This category of tool has expanded rapidly across the marketing technology sector over the past several months. PPC Land reported in December 2025 that impact.com had partnered with Evertune to integrate AI brand-monitoring reports into a performance marketing platform, while Amplitude and Adobe launched comparable visibility and optimization tools in October and November 2025 respectively. Not every industry voice regards the category as durable: the founder of one AI-visibility tool, Lorelight, shut the service down in October 2025 after concluding that brands with strong AI search visibility shared the same underlying traits as those with strong traditional SEO performance - quality content, third-party authority, and genuine reputation - rather than any AI-specific technique.

AI Overview share climbs to roughly 35 percent of German searches

The newsletter's second-quarter product summary included a data point with direct relevance for the broader AI-search measurement debate. According to SISTRIX, the share of search queries triggering an AI Overview in Germany now stands at approximately 35 percent, with the trend continuing upward.

That figure extends a trajectory PPC Land has tracked closely since SISTRIX first began publishing German AI Overview coverage data. SISTRIX recorded the share at approximately 17 percent of German keywords in August 2025. By February 2026, the company's monthly review put the figure at approximately 20 percent of all keywords, a share the analysis tied to a cost of 265 million lost organic clicks per month across the German market, alongside a click-through-rate collapse at the top organic position from 27 percent to 11 percent when an AI Overview appeared on the page. The March 2026 review reaffirmed those figures. The July newsletter's approximately 35 percent figure represents roughly double the share recorded eleven months earlier.

SISTRIX connected the July figure to a separate development: reports, gathered from online forums, of visibility and traffic shifts at some websites around July 11, 2026, that several observers initially interpreted as an unconfirmed Google algorithm update. According to the newsletter, SISTRIX's own measurements suggest those shifts are more likely attributable to increased AI Overview display frequency than to changes in underlying organic rankings - a conclusion consistent with the company's own recorded rise in AI Overview prevalence over the same period.

Second-quarter product roadmap adds Prompt Research, AI Deep Dives and open MCP access

The newsletter's final major section detailed SISTRIX's product development across the second quarter of 2026, framed around the stated goal of making analysis and measurement for AI-driven search "better and easier to implement" as the channel's share of search continues to grow.

Prompt Research, according to SISTRIX, consolidates more than 62 million real user queries into more than 1.4 million distinct topics, providing what the company describes as a systematic picture of demand within AI search, including search volume, intent classification and a ready-made content brief for each topic. PPC Land reported on the original launch of Prompt Research in SISTRIX's April 2026 newsletter, where the tool was introduced as a structural replacement for keyword-based research methods in an environment where user intent is expressed in full sentences rather than short queries.

AI Deep Dives allow SISTRIX data to be interpreted and analyzed directly within the tool by AI systems, with results calculated deterministically from the company's own data rather than generated freely, and available for any tracked brand rather than only a user's own domain. Eleven Deep Dives are available at launch, covering subjects such as brand perception and content opportunity analysis. PPC Land covered the launch of AI Deep Dives when SISTRIX first introduced the eleven-dive set, distributed across the AI Check and Prompt Research products, with four of the eleven dives addressing content-opportunity and buyer-type analysis within Prompt Research specifically.

The MCP Server, according to the newsletter, now operates through OAuth authentication and is available to all SISTRIX users without requiring an API key or consuming API credits, allowing SEO tasks to be carried out directly within Claude or ChatGPT using SISTRIX's underlying data. PPC Land reported on the original OAuth-based access expansion on May 21, 2026, noting that the change removed a credential requirement that had previously restricted the capability to Plus-tier subscribers and above. A further free webinar covering setup and use cases, hosted by Julia Weißbach, is scheduled for September 9, 2026.

SISTRIX said additional detail on how these tools combine in day-to-day marketing workflows, along with other second-quarter developments including new API functions for domain and directory data, appears in a detailed, illustrated quarterly review published on sistrix.de.

The newsletter also flagged three additional items relevant to search marketers, each attributed to reporting from seo-suedwest.de rather than to SISTRIX's own research.

The first concerned the unconfirmed Google update discussed above. The second described a new Platform Propertiesfeature inside Google Search Console that lets site owners track how their Instagram, TikTok, X and YouTube content performs in Google Search, including a performance report and insights, with the rollout proceeding gradually over the following weeks. PPC Land documented the launch of platform properties when Google introduced the feature, and separately reported on its global rollout to all accounts, noting the feature extended reporting logic Google had built for website properties to social and video accounts for the first time.

The third item concerned Google Preferred Sources, which the newsletter said became visible within German-language AI Overviews following the feature's global rollout in May 2026, with the potential for increased clicks and increased trust attached to labeled sources. According to the newsletter, whether the feature achieves lasting adoption remains unclear. PPC Land reported on the May 27, 2026 expansion of Preferred Sources into AI Overviews and AI Mode, when Google disclosed that users had selected more than 345,000 unique sources through the mechanism and that labeled links received roughly double the click-through rate of unlabeled ones.

Why this matters for the marketing community

The July newsletter connects two developments that, taken separately, might seem unrelated but together describe a single shift in how search-visibility work is measured. On one side sits a maturing set of tools for observing how AI systems already talk about a brand - sentiment, competitive positioning, and recommendation presence - built, according to SISTRIX, entirely from documented real-world AI responses rather than modeled projections. On the other sits evidence that the quality of AI-generated content itself remains highly variable, and that variability appears driven primarily by how a piece of content is instructed to be written rather than by which underlying model produces it.

For a publication like PPC Land, which has tracked SISTRIX's monthly data releases as one of the more consistent empirical sources on the German search market, the July figures extend rather than reset the picture. The approximately 35 percent AI Overview share reported this month sits within a trajectory that moved from roughly 17 percent in August 2025 to roughly 20 percent by February 2026, a progression documented across successive PPC Land reports on SISTRIX's monthly reviews. That the newsletter also cited this same rising share as the likely explanation for a wave of unconfirmed-update speculation illustrates a recurring pattern in the current search landscape: traffic and visibility shifts that once would have been attributed reflexively to a ranking algorithm change increasingly trace back instead to the growing footprint of AI-generated answers displacing organic results before a click ever occurs.

The brand-perception tools, meanwhile, arrive into a market segment that has grown quickly but remains unsettled on questions of durability and methodology. PPC Land's coverage of comparable launches from Evertune, Amplitude and Adobe throughout late 2025 documented rising vendor interest in AI-visibility measurement, even as at least one practitioner in the space concluded publicly that no AI-specific optimization technique exists beyond conventional content and reputation fundamentals. SISTRIX's own framing - that every output in its new analyses traces to a specific, citable AI response rather than an aggregated or inferred score - positions the tool within the more evidence-grounded end of that spectrum, though the newsletter does not disclose sample sizes, response collection dates, or how frequently the underlying data refreshes.

Timeline

Summary

Who: SISTRIX GmbH, a Bonn-based search analytics company founded and led by Johannes Beus, published the findings in partnership with content-marketing firm Wortliga for the AI text-quality study.

What: SISTRIX released data showing that AI-generated B2B texts from Claude, Gemini and ChatGPT average 44.1 points on a 100-point quality scale, with ChatGPT scoring lowest at 37.7 points and Claude highest at 47.7 points, and found that prompt design causes far larger score swings than model choice. Separately, SISTRIX introduced three new brand-perception analyses - Sentiment Analysis, Competitive Perception and a Recommendation Map - inside its AI Check tool, and reported that AI Overviews now appear on approximately 35 percent of German search queries.

When: The newsletter was published on July 31, 2026, summarizing research and product developments from across the second quarter of 2026.

Where: The findings concern the German search market specifically for the AI Overview share data, while the AI text-quality study and the new AI Check analyses apply across the AI platforms and brands they measure without a stated geographic restriction.

Why: The findings matter because they document, with specific figures, two separate but related shifts reshaping search marketing: AI-generated answers are displacing an increasing share of traditional organic clicks in Germany, and the content marketers increasingly rely on AI systems to produce carries substantial, prompt-dependent quality variation that current classification tools have only begun to measure systematically.