March 23, 202616 min read

Sentiment in AI: Which Brands Neural Networks Praise and Criticize — GEO Scout Research

Research on positive and negative rate of brands in neural network responses. Which brands AI praises (Timeweb 94%, Ozon 92%) and which it criticizes (Airbnb 12% negative). GEO Scout data across 5 niches and 9 AI providers.

Vladislav Puchkov

Founder of GEO Scout, GEO optimization expert

When a neural network answers "which marketplace is best" or "where to learn programming," it doesn't simply list names. It forms an opinion. Some brands get enthusiastic descriptions — "reliable," "best choice," "recommended." Others receive cautious caveats, warnings, or direct criticism.

We researched brand mention sentiment across 5 niches using GEO Scout data from March 2026. 9 AI providers: ChatGPT, Claude, DeepSeek, Gemini, Google AI Mode, Google AI Overview, Grok, Perplexity, and YandexGPT. The results show: sentiment in AI follows clear patterns, and understanding these patterns is the key to managing brand reputation in the age of AI-powered search.

What are positive rate and negative rate in AI

Before diving into the data, let's define the metrics. When AI mentions a brand in a response, the mention can be:

• Positive — AI recommends, praises, names advantages. Example: "Timeweb is an excellent hosting choice, with stable uptime and responsive support"
• Neutral — AI simply lists the brand without assessment. Example: "Among hosting providers, you can consider Timeweb, Beget, REG.RU"
• Negative — AI criticizes, warns about problems, points out drawbacks. Example: "Airbnb is blocked in Russia, use alternatives instead"

Positive rate — the percentage of responses where the brand is mentioned in a positive context, out of all responses where it appears. Negative rate — the same, but for negative context. The difference to 100% is neutral mentions.

The formula is straightforward:

Important: a high positive rate doesn't mean the brand is frequently mentioned. TripIt has a positive rate of 96.55%, but its mention rate is significantly lower. Sentiment and mention frequency are independent metrics.

Top 11 most positive brands in AI

According to GEO Scout data, here are the brands with the highest positive rate across all researched niches:

Several observations.

Zero negativity for 9 of 11 leaders. Almost all top brands receive no negative mentions at all. AI either praises them or mentions them neutrally. The only exception is Ozon with a minimal negative rate of 0.46%.

Small brands outperform giants. TripIt with 96.55% positive is a niche travel planning service. Zolotoye Yabloko at 88.89% is far from the largest e-commerce player. Yet AI loves them. The reason: for smaller brands, the mention context is narrower and more positive. AI mentions them only when they're truly relevant — and in those cases, it praises them.

Hosting dominates. Timeweb (94%) and Beget (89%) — two hosting brands in the top 5. This isn't a coincidence but a systemic pattern, which we'll discuss below.

Brands with negative sentiment: who AI criticizes

Negative sentiment in AI is rare. Most brands have a negative rate of 0%. But there are exceptions, and they form a clear pattern:

The most striking case is Airbnb. Positive rate and negative rate are identical: 12.24% each. This means that for every positive mention, there is exactly one negative one. The remaining 75.52% are neutral. For a globally recognized brand, this is a catastrophic sentiment profile.

Wildberries is the only one of the top 3 Russian marketplaces with a noticeable negative rate (1.92%). For comparison: Ozon — 0.46%, Yandex.Market — 0%. AI captures negativity related to the marketplace's conflicts with sellers, changing terms, and buyer complaints. At the same time, Wildberries has a high positive rate — 80.29%. The brand is simultaneously praised and criticized.

AliExpress (negative 3.20%) receives criticism for long delivery times and product quality concerns. AI often adds caveats like "delivery may take several weeks" or "pay attention to seller reviews," which registers as negative context.

Blocked services: how geopolitics shapes AI sentiment

The strongest factor in negative sentiment isn't a bad product or customer complaints. It's service blocking. GEO Scout data is unequivocal:

The pattern is clear: complete blocking kills sentiment, while partial restrictions reduce it. Airbnb and Booking in the Russian-language context receive more negativity than positivity, or comparable levels. AI providers trained on Russian-language sources carefully warn users: "the service is unavailable," "consider alternatives," "there are access issues from Russia."

This creates a vicious cycle. AI generates negative context. New data with this context enters the training set. The next AI version amplifies the negativity. For blocked brands operating in the Russian market through workarounds, this is a serious problem — neural networks actively discourage users.

At the global level, the situation is entirely different. Airbnb and Booking are among the strongest brands in worldwide AI search. But in Russian-language responses, their sentiment is radically different. This clearly shows that sentiment in AI is localized: the same brand can be a "favorite" in one language context and an "outsider" in another.

By niche: which sector does AI praise most

We calculated the average positive rate of leaders in each of the 5 researched niches. The results form a clear hierarchy:

Hosting is the absolute leader in sentiment. Timeweb (94.09%), Beget (89.19%) — this isn't just a high positive rate, it's practically a complete absence of neutral mentions. When AI talks about hosting providers, it doesn't just list them — it recommends them. The reasons:

• Hosting is a technical category with measurable characteristics (uptime, speed, price). AI can easily evaluate objective parameters
• Hosting providers have an enormous base of technical content: documentation, tutorials, reviews. This content is predominantly positive
• Hosting has few emotional reviews — unlike EdTech or e-commerce, harsh "takedown" reviews are rare here

Travel ranks second (~84%), but with an important caveat: this niche also contains brands with the highest negativity (Airbnb, Booking). Travel is a niche of contrasts. Local brands (Yandex Travel, Aviasales) receive steady positivity. Internationally blocked services get harsh negativity.

FinTech is stable (~80%). Alfa-Bank (84.13%), T-Bank (84.07%) — AI praises digital banks for innovation and convenience. Negativity is virtually absent. But there's no extreme positivity either — FinTech doesn't inspire the same "enthusiasm" from AI as hosting.

Why EdTech is the least positive sector in AI

EdTech shows an average positive rate among leaders of about 68% — 22 percentage points below hosting. That's a massive difference. Let's examine why.

Data on key EdTech brands

Three reasons for low positive rate in EdTech

1. Subjectivity of reviews. In hosting, uptime is a number. In education, "course quality" is an opinion. The internet is full of contradictory reviews about every EdTech platform: some are delighted, others are disappointed. AI aggregates this ambiguity and takes a cautious stance. Instead of "Netologia is an excellent choice," the neural network writes "Netologia is one of the options, pay attention to reviews."

2. Student complaints. Educational platforms generate a significant volume of negative user content: forums, social media, review sites. GeekBrains (negative rate 0.66%) is a prime example. Complaints about program changes, teaching quality, and refund difficulties — all of this enters AI training data. For comparison: Timeweb (hosting) has minimal such content.

3. The free platform phenomenon. freeCodeCamp (81.25%) and HubSpot Academy (84.21%) receive higher sentiment than paid competitors. AI is more positive toward free educational resources — the absence of a commercial component removes an entire layer of potential negativity (price, refunds, unmet expectations).

For EdTech companies, this means: managing sentiment in AI requires systematic work with reviews and creating positive content that outweighs negative user experiences.

Neutral brands: aggregators and informational sites

A separate category — brands with anomalously low positive rate and zero negativity:

Two subtypes stand out here.

Aggregators: AI doesn't praise information sources

HostingHUB has a positive rate of 0% — yet its domain citation is 49.39%. AI actively uses HostingHUB content as a source but doesn't recommend the brand itself. Similarly, Tutortop: positive rate 2.94%, but AI cites the site when forming EdTech responses.

This is a fundamental insight: AI separates "information source" from "brand recommendation." Aggregators and informational sites receive citations but not positive sentiment. Their role in the AI ecosystem is to supply facts, not to receive recommendations.

Low-awareness brands: no data — no sentiment

Kuper (0% positive) and DOM Bank (0% positive) are brands with low mention rates. When AI mentions them, it does so neutrally — it simply doesn't have enough data to form an opinion. No reviews, no expert content — no sentiment.

For such brands, the task isn't "improve sentiment" but first create a content base from which AI can form a positive opinion.

Five sentiment patterns in AI

Based on GEO Scout data across 5 niches, we identify five consistent patterns:

1. Technical infrastructure receives maximum positivity

Hosting is the most positive sector. AI praises what can be measured: 99.9% uptime, loading speed, pricing. The more objective a product's characteristics, the more confidently AI recommends it.

2. Service blocking = negative sentiment

Services blocked in Russia receive record negative rates. Airbnb (12.24% negative) is the most striking example. AI actively warns about unavailability, which is recorded as negativity.

3. Smaller niche brands are often more positive than giants

TripIt (96.55%), Zolotoye Yabloko (88.89%) — niche brands with local audiences. Their positive rate is higher than the largest market players. The reason: AI mentions them in a narrow, relevant context where sentiment is inherently more positive.

4. Aggregators have zero positive rate

HostingHUB (0%), Tutortop (2.94%) — AI doesn't praise information sources. It uses their content but doesn't form a positive attitude toward the brand itself. Citation and sentiment are different dimensions of AI visibility.

5. EdTech is a zone of heightened AI caution

In education, AI takes a neutral position. The average positive rate among leaders is 68%, significantly lower than other sectors. The reason: an abundance of subjective reviews, complaints, and contradictory opinions in training data.

How to improve brand sentiment in AI: a practical guide

Sentiment in AI is not random. It's the result of what content about your brand is available to neural networks. Here are concrete steps to improve your positive rate.

Working with content

Publish case studies with measurable results. AI responds best to specifics. Not "we improved our client's metrics," but "uptime grew from 99.5% to 99.99%, loading speed decreased from 3.2 to 0.8 seconds." This is exactly why hosting brands get such high positive rates — their content is packed with measurable characteristics.

Create expert content, not promotional content. AI distinguishes useful content from promotional material. Tutorials, documentation, research create a positive context. Advertising landing pages don't.

Use structured data. FAQ schema, HowTo schema, Review schema help AI correctly interpret your site's content. More on technical optimization in the article about GEO site audit.

Reputation management

Address negative reviews. Don't delete them — respond to them. AI considers context: if a negative review has a professional response with a solution, it neutralizes the negativity.

Encourage positive reviews on independent platforms. AI values reviews on your own site less than reviews on third-party resources. Reviews on themed platforms, mentions in rankings — all of this shapes positive rate.

Monitor sentiment by provider. Different AI systems may show different sentiment for the same brand. GEO Scout lets you see positive and negative rates for each of 9 providers — and find specific points of negativity.

Strategic actions

For EdTech: focus on graduates, employment outcomes, specific skills. Create content that outweighs negativity from review sites.

For e-commerce: publish comparative reviews with objective data (prices, delivery times, warranties). AI values facts, not marketing promises.

For FinTech: share information about technological innovations and user experience. AI responds positively to digital transformation.

Recommendations: what to do with this data

For brands with high positive rate (80%+)

Your task is retention. Continue creating expert content, maintain product quality. Monitor dynamics through GEO Scout — positive rate can decline if a wave of negative reviews appears online.

For brands with negative rate > 0%

Find the source of negativity. In most cases, it's a specific issue: service blocking, mass complaints, a scandal. Address the root cause, not the symptoms. If the reason is blocking, create content about legitimate alternative ways to use the service.

For brands with zero positive rate

You need to first build a content foundation. AI cannot praise a brand about which there is no positive information. Start with publishing expert content, case studies, and participating in independent rankings and reviews.

For EdTech companies

Focus on objective metrics: employment rate, average graduate salary, course ratings on independent platforms. AI is cautious with EdTech — give it objective data to form a positive opinion.

For everyone

Sentiment monitoring should be regular. Sentiment in AI changes — new reviews, model updates, context changes can sharply shift positive and negative rates. The GEO Scout platform lets you track these changes daily across all 9 AI providers. Getting into neural network recommendations is the first step, but maintaining positive sentiment is an equally important task.

Methodology

Data was collected by the GEO Scout platform on March 23, 2026. Monitoring covers 9 AI providers: ChatGPT, Claude, DeepSeek, Gemini, Google AI Mode, Google AI Overview, Grok, Perplexity, and YandexGPT (Yandex with Alisa). The research covered 5 niches: hosting, travel, e-commerce, FinTech, EdTech. Sentiment is determined automatically based on analysis of the brand mention context in the AI response. Positive rate and negative rate are calculated as the proportion of positive/negative mentions out of the total responses in which the brand appears.

Full interactive rankings for all niches are available on the GEO Scout ratings page. To monitor your brand's sentiment in real time, register at geoscout.pro. The Command Center at GEO Scout automatically analyzes sentiment for each provider and generates specific tasks — what content to create to improve how neural networks perceive your brand.