ChatGPT ads cluster around shopping questions in Australian insurance study

Somantra says a study of 34,279 ChatGPT insurance queries in Australia found ads appear more often around shopping-style prompts than explanation or claims questions. The findings suggest brands may gain visibility when buyers are forming a shortlist, but wording changes alone do not fully explain ad placement.

Why it matters: - Somantra's findings suggest ChatGPT ad placement is not random in Australian insurance queries. - Shopping-intent prompts appear more likely to surface ads, which could affect how brands show up when consumers are comparing cover. - The pattern matters for insurers and marketers trying to win attention at the point of consideration.

What happened: - Somantra analysed 34,279 ChatGPT insurance queries from Australia in August 2026. - The Sydney-based AI search brand visibility platform said queries using words such as "best", "top" or "recommended" carried an ad 85% of the time, versus 76% across all queries. - Explanation and claims questions showed fewer ads than shopping-style prompts.

The details: - Somantra said responses with shopping language were more likely to include ads by 8.0 percentage points when queries used "best", "top" or "recommended". - Queries mentioning a state or city were 5.2 points more likely to carry an ad. - Queries using "discount", "deal" or "save" were 2.3 points more likely to carry an ad. - Questions starting with "why" were 6.8 points less likely to show an ad. - Questions using "what is" or "what does" were 6.0 points less likely to show an ad. - Claims-related questions were 5.1 points less likely to show an ad. - Even so, "why" questions still carried an ad 72% of the time. - Words such as "quote", "price" and "buy" showed no clear link to ads. - Somantra audited wording effects on two separate days in August 2026. - The company also analysed 31,820 ChatGPT responses after removing repeated response text.

Between the lines: - The data points to a high base rate of ads, with wording shifting the odds around the edges rather than fully determining placement. - Somantra said the same question asked twice produced a different ad outcome 31% of the time across 747 pairs. - That level of variation means wording changes have to outperform normal randomness to show a real effect. - Adding "best" raised the ad rate from 82.5% to 96.9% across 97 pairs, with an ad appearing in 17 pairs and disappearing in 3. - Adding "cheap" raised the ad rate from 67% to 91% across 33 pairs. - In paired tests with and without "cheap" or "cheapest", Budget Direct's ad appeared only on the version with the word in 23 pairs and only on the version without it in 4. - "Online" and "commercial" showed the same pattern. - Somantra said all reported differences passed its statistical checks, but the results do not prove wording caused an ad to appear. - The company also said the tested query features explain little of the overall variation. - The one-word comparison covered 281 pairs, and several words had fewer than 20 examples.

What's next: - Somantra is framing the findings as early evidence of how advertising may be shaping visibility inside ChatGPT in Australia. - The company says brands can use the pattern to understand when paid placement is more likely to appear in shopping journeys. - The full report is available free in Somantra's report, and brands can request a free AI search brand audit.

The bottom line: - In Somantra's view, ChatGPT ads tend to move in when users shop and recede when users ask how or why insurance works.