Resources · Fixing Wrong AI Answers

How Can a Financial Advisor Check Whether Approved Client Testimonials Make Gemini Overstate Investment Results?

A financial advisor can check whether approved client testimonials make Gemini overstate investment results by testing it with a fixed set of questions. Ask about returns, results and reviews in Gemini and Google's AI features, and capture each answer with its sources. Then score each answer. Does it state a return figure? Does it attribute the figure to one client, or present it as typical or expected? Does it keep the context your testimonial page gives? Trace every overstated figure back to the page it came from, and change that page with your compliance team. Put disclosures next to each quote in text, remove stray figures, and present performance properly on its own page. An approved testimonial can still read like a performance claim once summarised.

Why can an approved testimonial become an overstated answer?

An approved testimonial page is reviewed as a whole: the quote, the heading, the disclosures and the page layout. Gemini does not read it that way. It extracts sentences and combines them with other sources. Common patterns:

  • A figure travels without its context. "My retirement account grew 38% in two years" becomes "clients have seen returns of up to 38%".
  • One client becomes all clients. Three positive quotes become "clients consistently report strong returns".
  • Disclosures are left behind. Statements that the person is a current client, whether they were paid, and that results are not representative often sit in a footer, a collapsed section or an image. Those are the parts least likely to be carried into a summary.
  • Testimonials mix with market figures. A quote about "great growth" sits near a chart of market returns, and the two get merged.
  • Third-party reviews add numbers. Reviews on other sites may mention figures your compliance team never approved, and answers draw on them too.

What does the SEC Marketing Rule require?

For SEC-registered advisers, the Marketing Rule sets the standard your own pages are held to. An advertisement may not include any untrue or misleading statement. It may not include information likely to cause a misleading implication about the adviser. It may not discuss potential benefits without fair and balanced treatment of material risks or limitations. It may not include or exclude performance results, or choose time periods, in a way that is not fair and balanced.

A testimonial must be accompanied, clearly and prominently, by disclosures stating:

  • that it was given by a current client or investor;
  • whether cash or non-cash compensation was provided for it; and
  • any material conflicts of interest arising from the person's relationship with the adviser.

Advertised performance has its own conditions. Gross performance must be shown with net performance, with at least equal prominence. Performance results for portfolios other than private funds must include one-, five- and ten-year periods, presented with equal prominence. Hypothetical performance is subject to further restrictions.

Gemini's answer is not your advertisement. But if Gemini draws a misleading impression from your testimonial page, compliance should ask whether the page itself invites that reading. And if your firm quotes or shares an AI summary in its marketing, treat that summary as your own communication. FINRA's rules for broker-dealers, which govern third-party content a firm adopts or helps prepare, are covered in how a compliance officer can tell whether an AI citation comes from an unapproved webpage.

Which questions should you test?

Use the questions prospective clients ask, including the leading ones:

Question What it tests
"What returns do [firm] clients get?" Whether testimonial figures are reported as typical
"Is [firm] good at growing retirement savings?" Whether benefits are stated without limits
"[Firm] reviews" Which reviews and testimonials are summarised, and how
"Can [firm] beat the market?" Whether answers imply outperformance
"How much will my portfolio grow with [firm]?" Whether answers project results
"Best financial advisor in [city] for retirees" How your firm is described among competitors
"What do clients say about [advisor name]?" Whether individual quotes are carried with context

Run each question three times in Gemini and in Google's AI features, and once or twice in ChatGPT and Perplexity for comparison. How to capture Gemini answers consistently is covered in how to check what Gemini says about your business.

How do you score each answer?

Record for every answer:

  1. Figure stated: does it give a return, growth or performance figure? Copy it exactly.
  2. Source of the figure: which cited page contains it: a testimonial, a performance page, a third-party review, or no visible source.
  3. Attribution: is it attributed to one named or described client, or presented as the firm's result?
  4. Typicality: does the answer present it as typical, expected, likely or guaranteed?
  5. Context kept: does it mention that results vary, that past performance does not guarantee future results, or that the speaker was a client?
  6. Overstated: yes if the answer presents a client's figure as typical or expected, projects results, or states performance your approved materials do not support.

Then calculate an overstatement rate: overstated answers divided by all answers. For example, if 60 answers include 11 that present a testimonial figure as typical, the rate is 18%. Track it by question and by source page, so you can see which page drives the problem. How regulated firms can measure AI visibility without making promotional claims in the process is covered in how regulated companies can measure AI visibility.

How do you fix the source page?

Work through the pages that produce overstated answers, with compliance approving each change:

  • Put disclosures in text next to each quote. Place client status, compensation and conflicts directly under the testimonial, not in a footer, tooltip or image.
  • Reconsider figures in testimonials. A quote about service, communication or planning is less likely to be turned into a performance claim than one with a percentage.
  • Label the section plainly, for example "What some clients say about working with us", rather than "Client results".
  • Keep performance on its own page, presented as the rule requires, so answers asked about returns have a proper source to cite.
  • Separate testimonials from charts and market commentary.
  • Check third-party review sites for figures you did not approve. Whether and how you can respond depends on your involvement with those sites, so involve compliance.

After changes, rerun the same questions. Where Gemini keeps producing a misleading figure, use the thumbs-down feedback and describe the error. How Google handles correction requests is covered in whether you can ask Google to correct a false Gemini answer.

What might an illustrative case look like?

Consider an illustrative registered investment adviser with a testimonial page of eight quotes. Two mentioned percentages, and the disclosures sat in a footer. In 60 answers across Gemini and Google's AI features, 11 said clients "have seen gains of up to 41%" or "typically outperform". All 11 cited the testimonial page. Compliance approved a revised page: the two percentage quotes were replaced with quotes about planning, disclosures were placed under each quote, and the section was retitled. A separate performance page gave net figures for one, five and ten years. Six weeks later, 2 of 60 answers still repeated the old figure, both citing a third-party article that quoted the old page.

What do people ask about testimonials and AI answers?

Are we responsible for what Gemini says?

Its answers are not your advertisements, but your pages are. Keep them accurate and hard to misread, and do not reuse AI summaries in your own marketing without review.

Should we remove all testimonials?

Not necessarily. Testimonials without figures, with clear disclosures next to them, are much less likely to be overstated.

How often should we test?

Quarterly, and after any change to testimonial or performance pages.

Published by The Kovmere Team · updated Sep 29, 2026 · More resources