The short version
A practical consumer guide to comparing 20 products with AI without getting fooled by fake reviews. Covers the verified fake-review numbers, what Amazon, Google, ChatGPT, and Perplexity shopping features actually do, the review red flags regulators endorse, and an eight-step workflow that keeps humans in charge.
Summarize with AI
Editorial Disclosure & Affiliate Notice
This content is published for informational and educational purposes only. It is not intended as a substitute for professional, legal, financial, or medical advice. AIUnpacker is funded through clearly labeled sponsored articles and reviews, relevant paid link placements, and affiliate commissions where disclosed. Commercial partnerships help fund the free editorial library. Sponsored work is labeled, and a payment does not change the verdict.
- For educational purposes only. Nothing here should be taken as a guarantee, recommendation, or professional recommendation.
- AI-assisted editing. Drafts are produced with AI assistance and get a human pass before they go up.
- Opinions are our own. Also, we are not affiliated with most tools we cover unless explicitly stated.
- Information may be outdated. Verify pricing, features, and policies directly with the vendor.
- Last reviewed: . Published .
Read more on our About page, Terms and Editorial Policy.
You can compare 20 products with AI in about an hour, as long as the AI never gets the final say. The pattern that works: lock in verified specs from manufacturer pages first, then use AI to compress each product’s reviews into themes, cross-check those themes against the specs and independent expert reviews, screen every product for fake-review signals, and spot-check the finalists yourself before buying. The reason for the human checkpoints is simple. An AI shopping assistant summarizes whatever reviews it’s fed, and in the UK government’s research, 11-15% of reviews in common product categories are likely fake. Summarize fake reviews and you get a confident, polished summary of lies.
That’s not a reason to skip AI. It’s a reason to give it guardrails.
Why fake reviews poison AI product comparison
AI comparison tools don’t verify anything. They compress. If a chunk of a product’s reviews were bought, an AI summary will fold them into its “consensus” and hand you a buyer’s guide that reads beautifully and lies fluently.
The scale of the pollution is measurable. The UK’s Department for Business and Trade published research in April 2023 estimating that 11-15% of reviews on e-commerce platforms in consumer electronics, home and kitchen, and sports and outdoors are likely fake, and that shoppers were 9.2% more likely to buy a product with well-written fake reviews once its price passed £80 (GOV.UK). The CMA cited research suggesting up to 50% of online reviews may be fake when it opened its first fake-review investigations in March 2026 (GOV.UK).
Platforms are filtering hard. Trustpilot removed 4.5 million fake reviews in 2024 (7.4% of everything submitted that year) with over 90% caught automatically (Trustpilot Trust Report), and Yelp filtered out nearly 500,000 suspected AI-generated reviews in 2025 (Yelp Trust & Safety Report). Amazon says it blocked more than 250 million suspected fake reviews in 2023 alone (aboutamazon). Filtering helps, but new fakes appear daily; the FTC’s first rule-based warning letters went to ten businesses in December 2025 (FTC Business Blog).
Between 11% and 15% of reviews in common product categories are likely fake. UK government research, April 2023. An AI that “reads all the reviews” reads the fake ones too, and summarizes them with equal confidence.
Adoption is running ahead of trust, which is the right instinct. Alchemer’s August 2026 survey found 48.5% of US shoppers had used AI to research a purchase in the past year, but only 35.4% trust AI recommendations (Alchemer). Use AI for speed; keep the skepticism.
Fakespot is gone: the review-checker situation in 2026
Fakespot no longer exists, so older advice to “run it through Fakespot” is dead. Mozilla announced the shutdown on May 22, 2025 and pulled the site, extensions, mobile apps, and Firefox’s built-in Review Checker on July 1, 2025 (Mozilla blog). ReviewMeta, the other well-known Amazon review analyzer, went offline in mid-2024. A few smaller tools have tried to fill the gap; treat their grades as one signal at most.
A review checker is a tool that grades the authenticity of a product’s reviews rather than the product itself. With the mainstream ones gone, the replacement is behavioral: screen for the review-fraud signals in the next section and anchor everything in first-party specs and expert sources: roughly what Fakespot was approximating anyway.
Meanwhile, the AI assistants have moved into the gap. Amazon’s Rufus is reportedly being folded into an “Alexa for Shopping” experience during 2026, Google shipped shopping features inside AI Mode in Search, OpenAI launched shopping research in ChatGPT in November 2025, and Perplexity has run a full shopping experience since November 2024. None of them verifies that the underlying reviews are real, that part is still your job.
The fake-review signals that actually hold up
Only a handful of red flags survive contact with regulators’ guidance, and the FTC’s consumer advice lists most of them: review bursts, single-product reviewer accounts, suspiciously glowing language, and undisclosed incentives (FTC Consumer Advice). Run this checklist per product:
| Signal | What a fake pattern looks like | How to check |
|---|---|---|
| Review timing | A burst of reviews over a short period, then silence | Sort reviews by date and look for clusters |
| Reviewer history | Accounts created to review one product | Click through to the reviewer’s other reviews |
| Star distribution | A five-star wall with almost no mid or low ratings | Check the rating histogram; real products spread |
| Language | The same phrases repeated; fluent but generic | Search the page for duplicated sentences |
| Specificity | Praise with zero detail about actual use | Real reviews mention use cases, comparisons, flaws |
| Sentiment direction | Suspiciously glowing, or fake negatives attacking competitors | Read the critical reviews, not just the top ones |
| Incentives | Five-star reviews swapped for discounts, undisclosed | Look for incentive labels; undisclosed paid sentiment breaks the FTC rule |
| Purchase verification | Long detailed reviews from unverified buyers | Filter by verified purchase where possible |
Two nuances. Fake positives aren’t always five stars: the FTC notes some fakes give four stars to look credible, so “not all five stars” proves nothing. And “sounds real” is no longer a test: AI writes fluent prose by default, and Yelp’s filters alone caught nearly 500,000 suspected AI-written reviews in 2025. Fluency is free; specificity is not.
What Amazon, Google, ChatGPT, and Perplexity actually do
All four ecosystems now summarize shopping information with AI, and they fail in different ways. The current state, verified from each company’s own announcements:
| Tool | What it does | Key caveat |
|---|---|---|
| Amazon “Customers say” | AI paragraph plus clickable theme tags, built from Verified Purchase review text; live in the US, UK, Australia, India, Singapore, and UAE | Inherits fake-review pollution; sellers can’t correct errors |
| Google AI Mode shopping | Conversational discovery, virtual try-on, and price tracking launched May 2025; agentic checkout added November 2025 | Since October 2026, Search ads can show AI review lines drawn from positive feedback only |
| ChatGPT shopping research | Asks clarifying questions, researches the web, returns a cited buyer’s guide; rolling out to Free, Go, Plus, and Pro plans since November 24, 2025 | OpenAI itself says the model may get price and availability wrong: verify on merchant sites |
| Perplexity shopping | Unsponsored product cards, plain-language comparisons, and “Buy with Pro” checkout for US Pro subscribers; built on Shopify data since November 18, 2024 | Only as good as the indexed merchant data and reviews |
Amazon’s summary is the most constrained: highlights come only from Verified Purchase review text and need multiple customers sharing an opinion (aboutamazon). That filters for volume, not authenticity: a brokered batch of fake five-star reviews still reads like a consensus, and seller forum threads complain about inaccurate summaries no one can fix. If a highlight can’t be checked against the underlying reviews, ignore it. That is a judgment call, not a documented Amazon position.
Google’s newest move deserves suspicion: in late October 2026 it began rolling out AI-generated review summaries inside Search ads for ecommerce brands in the US and UK, built from web-wide reviews and described in early coverage as positive-only (Common Thread Collective). A paid ad quoting only the nice half of a dataset is marketing, not research.
ChatGPT’s shopping research is the most thorough for a shortlist: it asks about needs and budget, researches the web, and returns a cited guide, working best in detail-heavy categories like electronics and kitchen appliances (OpenAI). OpenAI’s own announcement says to verify price and availability on merchant sites. That’s the vendor confirming the spot-check step matters.
Perplexity runs the longest-standing dedicated shopping mode: unsponsored product cards, comparisons, and “Buy with Pro” checkout for US Pro subscribers (Perplexity). The inputs are still reviews and merchant feeds.
Usability research backs up the wariness: NN/g found people trust AI review summaries more when negatives and sources are included, and that summaries often omit the exact detail the shopper wanted (NN/g).
The 20-product workflow, step by step
Done right, this takes an hour of research plus an evening of verification. Order matters: specs before reviews, because a manufacturer spec sheet is the only input with no incentive to lie.
- Write the requirements brief. List must-haves, dealbreakers, budget cap, and how much each factor matters. Weight them before you’ve seen any marketing.
- Build the 20-product candidate list yourself, from two or three retailers, with exact model numbers. AI can suggest, but you keep the list: that’s where sponsored placements get excluded.
- Collect verified specs from manufacturer pages. Pull dimensions, materials, power, capacity, warranty, and safety certifications into a spreadsheet, straight from the maker’s site. If a page doesn’t state something, mark it unknown.
- Have AI extract review themes per product. Prompt it: “List the five most common positive and five most common negative themes, with an approximate count of mentions and verbatim phrases. If the negatives contradict the positives, say so.” Counts and quotes make a summary checkable.
- Cross-check the consensus against your spec table. If reviews say “surprisingly compact” and the spec sheet says 24 inches wide, someone’s wrong. This catches hallucinations too: invented review claims evaporate when you demand the quotes behind them.
- Screen each product for the fake-review signals from the table above. Bursts, five-star walls, duplicated phrasing, single-review accounts. Two signals on one product means discounting its review evidence and leaning on specs and expert sources instead.
- Build the comparison table and score it. Columns are your weighted criteria from step 1; rows are the 20 products; cells hold spec values, not adjectives. Let AI draft it, then audit a sample of cells against your spreadsheet.
- Verify the finalists independently. Check the top three to five against at least two expert sources (professional testers, hands-on reviews, community threads) plus a return-policy check. Buy the one that survives.
Skip steps 3 and 8 and the whole thing collapses. The verification steps are the difference between a comparison and a hallucination with a spreadsheet attached.
Where AI comparison still breaks down
AI product comparison fails in three documented ways, and every one of them has a workaround.
First, hallucinated consensus. Models can invent review themes or get prices wrong. OpenAI’s own announcement concedes the model may make mistakes on price and availability (OpenAI). The fix is mechanical: every material claim in your table needs a source you’ve opened.
Second, polluted inputs. Summaries inherit whatever the reviews contain, and Trustpilot’s and Yelp’s 2025 numbers show how much contamination survives filtering. The fix is step 6: screen the reviews before trusting their summary.
Third, commercial skew. Google’s ad-based review lines are positive-only by design, Amazon sellers can’t correct summaries they say are inaccurate, and the CMA opened five investigations into review practices in March 2026, including a review platform accused of leaving negative one-star reviews unpublished, and a company accused of undisclosed discounts for five-star reviews (GOV.UK). Treat anything a seller can influence as marketing, not evidence.
One legal footnote: since October 21, 2024, the FTC’s Consumer Reviews and Testimonials Rule has banned fake reviews, paid incentives conditioned on sentiment, and review suppression, with civil penalties up to $53,088 per violation (FTC Business Guidance, Federal Register). Enforcement is catching up (ten warning letters went out in December 2025) so don’t count on the law to filter your feed. Not legal advice, just the map.
What to do next
For quick checks, read the “Customers say” box, then the critical reviews and the spec sheet: three minutes, no tools. For a real 20-product comparison, run the workflow with whichever assistant you already use; ChatGPT shopping research is the strongest fit for cross-brand shortlists, Perplexity if you like unsponsored cards. Above a few hundred dollars, add professional testers (Consumer Reports in the US, Which? in the UK) as your tiebreaker, and report suspected fakes at ReportFraud.ftc.gov. AI earns its hour of savings only after you’ve given it verified inputs and kept the last word.
FAQ
Can AI reliably compare 20 products at once?
AI can reliably gather, organize, and summarize 20 products, but it can’t reliably judge them: most of its evidence is customer reviews, and a measurable share of those are fake. UK government research estimates 11-15% of reviews in common categories are likely fake. Treat AI output as a starting shortlist, anchor it with manufacturer specs, and verify the finalists before you buy.
Which AI shopping assistant is best for product comparison?
ChatGPT’s shopping research (launched November 24, 2025) is the most thorough for comparison because it asks clarifying questions and returns a cited buyer’s guide. Perplexity’s shopping mode produces unsponsored product cards. Amazon’s “Customers say” highlights and Google’s AI Mode shopping are faster but work product-by-product, and Google’s ad-based review lines are positive-only. The best choice depends on where you plan to buy.
How does AI-assisted product comparison work?
You define requirements and budget, build a candidate list with exact model numbers, and collect verified specs from manufacturer pages. Then AI summarizes review themes per product with mention counts and quotes, you cross-check that consensus against the specs, screen for fake-review signals, and build a weighted comparison table. The final step is independent verification of the winners.
How much does it cost to use AI for product comparison?
Nothing for the basics. ChatGPT’s shopping research is rolling out to Free, Go, Plus, and Pro accounts, and Amazon’s review highlights and Google’s AI Mode shopping are free. Perplexity’s “Buy with Pro” checkout needs a Pro subscription, but its shopping results work without one. Paid plans mainly raise usage limits; they don’t verify reviews.
What are the main alternatives to AI comparison tools?
Independent expert testers such as Consumer Reports or Which?, hands-on video reviews, retailer review filters, and community sources like Reddit. Fakespot shut down on July 1, 2025, and ReviewMeta went offline in mid-2024. Expert review sites and spec sheets remain the strongest counterweight to AI summaries, because they test products instead of paraphrasing reviews.
Which AI shopping tool is best for beginners?
Amazon’s “Customers say” highlights and theme tags are the easiest entry point: they sit on the product page you’re already reading and show clickable themes with positive and negative counts. ChatGPT shopping research is the better second step for a guided, cross-brand comparison. Both are free, and neither replaces the spec sheet.
What are the main limitations of AI product comparison?
Three documented ones. AI can hallucinate review consensus or get prices wrong. OpenAI’s own announcement says to verify price and availability on merchant sites. Inputs are polluted: fake reviews get summarized along with real ones. And commercial incentives skew output, as with Google’s positive-only review lines inside Search ads. AI summaries also tend to omit the detail you care about most (Nielsen Norman Group research).
Can sellers game AI review summaries?
Indirectly, yes. Summaries are generated from review text, so batches of purchased reviews can shape the “consensus” an AI reports. Sellers can’t edit an Amazon AI summary they believe is wrong: seller forum threads in 2025-2026 describe inaccurate ones. In the UK, the CMA opened five investigations into review practices in March 2026, including undisclosed incentives for five-star reviews.
How can users get better AI comparison results?
Give the AI structure. Specify must-haves and dealbreakers, provide model numbers, paste verified specs, and ask for review themes with approximate mention counts and verbatim quotes. Ask it to flag thin or contradictory evidence and include negatives: cited summaries that surface negatives are the trustworthy ones. Then spot-check the winners yourself.
What should users choose or do next?
Pick the tool that matches where you shop: Amazon highlights for quick single-product checks, ChatGPT shopping research or Perplexity for cross-brand shortlists. Run the eight-step workflow above for purchases over about a hundred dollars, keep the manufacturer spec sheet as your source of truth, and report suspected fakes to the FTC at ReportFraud.ftc.gov.
Sources
- Amazon: AI-generated review highlights (“Customers say”)
- OpenAI: Introducing shopping research in ChatGPT
- Perplexity: Shop like a Pro
- GOV.UK: Investigating the prevalence and impact of fake reviews
- GOV.UK: Fake and misleading reviews: 5 businesses under CMA investigation
- FTC: Consumer Reviews and Testimonials Rule: Questions and Answers
- Federal Register: Trade Regulation Rule on the Use of Consumer Reviews and Testimonials
- FTC Business Blog: Warning Letter or Ten: Comply with the FTC’s Consumer Review Rule
- FTC Consumer Advice. How To Evaluate Online Reviews
- Federal Register: Civil Penalty Inflation Adjustments (2026)
- Mozilla: Investing in what moves the internet forward (Fakespot shutdown)
- Trustpilot. Trust Report
- Yelp: 2025 Trust & Safety Report
- Amazon. Unmasking the fake review broker
- Alchemer: Half of shoppers use AI. Only a third trust it.
- Nielsen Norman Group. AI Summaries of Reviews
- Common Thread Collective: Google is now auto-generating review summaries in your Search ads
- Google: Agentic commerce tools for retailers and platforms
Weekly digest
Get our weekly AI digest
The latest AI tools, prompts, and insights — short, useful, every week.
AIUnpacker Editorial
Articles are written for a specific question. Sponsored work is labeled.
