A brand can appear in an AI-generated search answer and still lose the recommendation. That is the finding from Somantra, a Sydney-based AI search monitoring company, which analyzed 4,445 ChatGPT responses and found that adding a single decision-oriented word—such as “cheapest,” “safest” or “most trusted”—could substantially change how a brand was described without removing it from the answer.
For years, search marketers have focused on a relatively straightforward question: Does my brand appear when someone searches for it?
Generative AI is making that question considerably more complicated.
When a consumer asks ChatGPT to compare insurance providers, two nearly identical questions can produce very different descriptions of the same company. A brand might remain in the comparison table, but the reasons given for considering it can change dramatically.
A new study from Somantra is attempting to measure that phenomenon.
The Sydney-based company analyzed 4,445 ChatGPT responses to Australian insurance-shopping queries. From those results, researchers identified 135 matched query pairs where a single decision-oriented word was added to an otherwise identical query and the same brand remained present in both responses.
The methodology, known as perturbation testing, is designed to measure how sensitive AI-generated brand narratives are to small changes in user intent.
The findings suggest that AI search visibility is only one part of the equation.
Across the 135 observations, the median word shift was 0.94 on a scale from zero to one. In practical terms, Somantra says a typical modified query substantially rewrote the visible language used to describe a brand.
The median semantic distance was 0.53, indicating that the changes were not merely cosmetic. The underlying rationale for the recommendation often moved as well.
That distinction could become increasingly important for marketers optimizing for AI search.
Traditional SEO tends to revolve around rankings, impressions, clicks and keyword positions. AI search introduces another layer: how an AI system frames the company after deciding that it is relevant.
A brand can therefore win the visibility battle and lose the persuasion battle.
Price-related queries generated some of the largest changes in Somantra’s analysis. Adding terms such as “cheapest” or “most affordable” generated an average word shift of 0.97.
The word “best” produced the largest movement in reasoning, with a median semantic distance of 0.60. “Most trusted,” meanwhile, generated the largest positive sentiment movement, increasing net sentiment by 0.50 across 16 observations.
Those results point to an emerging challenge for AI search optimization, sometimes referred to as generative engine optimization or GEO.
Consumers rarely express purchase intent in exactly the same language. One person may ask for the cheapest insurance. Another may prioritize safety. A third may want the most trusted provider or the best coverage for a particular location.
A conventional search strategy might treat those phrases as closely related keywords.
An AI system can treat them as different decision frameworks.
Somantra’s research illustrates that difference through an Australian home-insurance example involving AAMI and QBE.
In the unmodified query, ChatGPT described AAMI as a large mainstream insurer providing home and contents coverage. After “safest” was added, the response reportedly shifted toward a more detailed value and coverage rationale involving Queensland performance and a named industry award.
QBE remained visible as well, but experienced a smaller change.
The significance is not that one insurer disappeared.
It is that one received a more compelling reason to be selected.
That distinction mirrors a broader change underway in search marketing.
Google’s search ecosystem has increasingly incorporated AI-generated summaries and conversational experiences, while Microsoft has integrated generative AI into Bing and Copilot. Meanwhile, businesses are experimenting with optimization strategies designed to influence how large language models interpret their brands.
The problem is that there is no equivalent of a traditional search-ranking position for an AI-generated narrative.
The same company can be mentioned in multiple responses while being characterized differently depending on the user’s stated objective.
Somantra’s study suggests marketers therefore need to monitor not only whether their brands appear, but what argument the AI makes for or against them.
That requires a more granular measurement framework.
The company uses word shift to measure how much the text changes, semantic distance to assess how much the underlying meaning changes, and sentiment movement to identify changes in positive or negative framing.
It also distinguishes perturbation testing from temporal monitoring.
Temporal monitoring examines how a brand’s visibility and citations change across repeated snapshots over time. Perturbation testing examines how the same brand responds to different formulations of essentially the same customer decision.
The two approaches answer different questions.
One asks: Is the AI’s treatment of my brand changing?
The other asks: How fragile is my brand’s positioning when customer intent changes?
That could matter considerably for marketing teams building AI-era content strategies.
A brand may have strong visibility for generic queries but weak positioning around price, trust or protection. Another company might appear less frequently overall but receive a stronger recommendation when customers use commercially important decision terms.
Home insurance produced the largest movement in Somantra’s research, with a median word shift of 0.98 across 46 clean observations. Roadside assistance, motorcycle insurance and travel insurance showed smaller but still measurable changes.
The category differences suggest that marketers may need to build query neighborhoods rather than monitor a handful of fixed prompts.
For an insurance brand, that could mean testing clusters around affordability, trust, coverage, claims experience, customer service and geographic suitability. For a software company, the equivalent questions might revolve around security, integrations, price, ease of deployment and enterprise scalability.
This approach also exposes a limitation of a single AI visibility score.
A visibility metric can tell a marketing team whether its brand appeared. It cannot necessarily tell the team whether the AI gave consumers a persuasive reason to choose it.
That is becoming a more important distinction as search behavior shifts from lists of links toward synthesized answers.
The findings should nevertheless be interpreted with some caution. The study is based on Australian insurance queries and ChatGPT responses, while Somantra is itself an AI-search monitoring company. The results demonstrate sensitivity within the tested dataset; they do not establish that every AI search engine or every category will behave identically.
Still, the underlying issue is broader than insurance.
As consumers increasingly use AI systems to research purchases, brand positioning inside the answer may become as important as brand inclusion.
For SEO and marketing teams, that means the optimization target is changing.
The question is no longer simply whether an AI model knows your company.
It is whether, when the customer’s priorities change, the model has a strong and accurate reason to recommend it.
Market Landscape
The emergence of AI search, generative search and GEO is changing how brands measure digital visibility.
Traditional SEO relies heavily on rankings, backlinks, structured content, technical optimization and user engagement. AI search introduces additional variables because large language models synthesize information and construct narratives rather than simply displaying ranked pages.
Platforms from Google, Microsoft and OpenAI are pushing search toward conversational and answer-oriented experiences. That creates a new optimization layer around:
- Brand inclusion — Does the company appear?
- Citation — Which sources does the AI rely on?
- Positioning — How is the company described?
- Reasoning — Why does the AI say consumers should consider it?
- Sentiment — Is the framing positive, neutral or negative?
- Intent sensitivity — Does the positioning change when customer priorities change?
For enterprise marketing teams, this means AI search measurement is likely to move from simple visibility tracking toward narrative and recommendation intelligence.
Somantra’s perturbation methodology is one example of that emerging measurement category.
Top Insights
- Somantra analyzed 4,445 ChatGPT insurance responses and found that single-word intent changes could substantially rewrite brand descriptions without removing brands from results.
- Price modifiers generated the largest textual changes, while “best” produced the greatest semantic movement across the Australian insurance queries tested.
- Home insurance showed the strongest sensitivity, suggesting marketers may need category-specific AI search strategies rather than relying on universal visibility scores.
- The study shifts attention from brand mentions toward recommendation quality, measuring whether AI-generated narratives provide persuasive reasons for consumers to choose companies.
- Perturbation testing complements temporal AI search monitoring by revealing how brand positioning changes when customer intent shifts within otherwise similar queries.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
