Do consumers trust AI recommendations? What it means for your brand's positioning in ChatGPT, Gemini and Perplexity

Almost one in two consumers trusts AI recommendations according to the chart attributed to NP Digital we examine here. Its methodology is unavailable, so we cannot extend that percentage to all consumers. It does invite consideration of ChatGPT, Gemini, Perplexity and Google's AI Overviews as reputation spaces as well as potential traffic channels. Accurate public information makes it easier for people or assistants to check your brand; it does not guarantee a recommendation.

Which sources do consumers trust most when evaluating a brand?

Short answer: people lead this chart. Friends and family (84%), customer reviews (69%) and independent experts (67%) take the top positions. AI recommendations sit in the middle at 48%, with advertising last at 5%.

Why is 48% trust in AI relevant?

In this comparison AI is close to brand websites and YouTube. That suggests an automated answer may be an initial filter for some people. Without a margin of error, a three- or four-point difference does not establish a statistically significant advantage.

It does not mean Reddit matters less to models. A forum may be cited in an answer even if the user never visits it. Separate stated trust in a channel from that channel's presence in AI answers.

The 48% does not establish that the remaining 52% distrusts AI: neutral or other response options may exist. Idea Grove provides a separate finding with a defined scope: its April 2026 Pollfish survey of 1,000 US adults reports that 98% check an unfamiliar brand recommended by AI. This measures verification, not 98% of the population buying after using an assistant.

How do AI models decide which brand to recommend?

A search-enabled assistant may consult pages, summarize them and cite some. Without search it responds from patterns learned during training and conversation context. It is not a complete directory and does not automatically confirm that its statements are true.

Reviews, articles, comparisons, forums and your website can provide public information. But the chart measures human trust: it does not reveal a selection formula or model weights. Our editorial recommendation is to use it to organize a reputation audit, not as a ranking algorithm.

For the practical side of getting cited, see our guide to GEO and appearing in ChatGPT. Here we focus on prioritizing sources and interpreting trust data.

Where should you invest to position your brand in AI models?

Compare the chart's sources with those appearing in queries about your sector. This gives a reasoned editorial priority, not a guarantee or a budget allocation suitable for every business:

1. Customer reviews (69%): request genuine feedback after an experience, not only from happy customers. Do not buy reviews or make rewards conditional on positive opinions. Reply to criticism too. Why: reviews help people verify brands and may supply public evidence.
2. Independent experts (67%): provide verifiable information to journalists, analysts and specialists for useful comparisons; respect their judgment and disclose paid collaborations. See this list of AI marketing experts for profiles to follow.
3. Google results (57%): maintain solid SEO, meaning useful content search engines can find and understand. Indexing supports discovery in Google, not guaranteed inclusion in every assistant.
4. Brand website (52%): answer real questions with prices, limitations, verifiable cases and identifiable people on your about page. It is your direct information source, although a model may not consult it.
5. Communities and Reddit (44%): participate helpfully without spam and disclose your connection to the brand. First check whether those communities appear in your sector's answers.
6. Friends and family (84%): support recommendations with a good experience and a transparent referral program. Incentives do not replace trust or automatically transfer it to AI.

Advertising, at 5% stated trust in this chart, may still help reach and conversion. The figure does not prove ads are useless or cannot build reputation. Support ads with genuine evidence such as reviews and cases, and measure effects separately.

Is appearing in ChatGPT enough?

Depending on one assistant is unwise. The chart's note says Gemini grew fivefold in two years but identifies neither the metric nor exact period: we do not use that statement as market-share evidence. The practical reason to diversify is that assistants may answer the same question differently.

• ChatGPT search: OpenAI says public websites may appear. Not blocking OAI-SearchBot, the program that consults pages, helps allow inclusion in summaries and snippets; it does not guarantee selection or citations.
• AI Overviews and AI Mode in Google Search: Google requires no special optimizations beyond ordinary SEO practices. That documentation should not be generalized to Gemini, a separate product.
• Perplexity: record the sources it shows and check whether they support its description of your brand. A citation alone does not establish a ranking criterion.
• Gemini, Claude and Copilot: review how they describe your brand if your audience uses them. Record whether search was used, the date and available settings.

Pew Research analyzed browsing by 900 US adults in March 2025: it observed result-link clicks on 8% of visits with an AI summary versus 15% without one. This is an observed association, not proof of causality or your website's CTR. CTR means click-through rate: clicks divided by impressions. See content strategies in the AI era.

The Garcai method: an initial 30-minute AI presence audit

You can run an initial review without coding. Thirty minutes is an indicative target, not a guaranteed duration; free access, service limits and availability can change:

1. Write five real customer questions; hypothetical example: ‘What's the best payroll software for small businesses in Spain?’.
2. Try ChatGPT, Gemini, Perplexity and Google. Use fresh conversations without history where possible; this does not remove all personalization. If no AI Overview appears, record that rather than forcing it.
3. Record date, model or service, search enabled, country, exact question, whether you appear, how you are described and cited sources. Do not treat list order as a stable ranking.
4. Classify sources: reviews, experts, communities, your website or others. Open each to check whether it actually supports the answer.
5. Find information gaps: few reviews, outdated comparisons, contradictory prices or incorrect facts. Prioritize what you can verify and correct.
6. Repeat comparable queries and record monthly trends. Changing one thing at a time supports learning, but this audit is not a controlled A/B test and cannot establish causality alone.

Common mistakes

• Looking for a keyword trick as if a public recommendation formula existed.
• Checking only ChatGPT without learning which assistants your audience uses.
• Removing advertising based only on a trust chart without methodology.
• Ignoring reviews and forums or attempting to manipulate them with fake opinions.
• Confusing a mention, citation and recommendation: these are different outcomes.
• Treating one answer as definitive: it may change by model, session and date.

No technique guarantees a model will recommend you. You can ensure people and assistants find reliable, current and consistent information, then use your own evidence to check what changes.