Reverse Engineering the Response to "best smartphones 2025" in Chatgpt 5.0

By Chris Byrne (18.8.25)

   This is my attempt to reverse engineer a response to a comparative entity prompt ("best smartphones 2025") in Chatgpt 5 using the tool itself. I chose this prompt to trigger "Retrieval Augmented Generation" (RAG): where instead of relying solely on the model’s pre-trained knowledge alone, the model retrieves relevant information from connected data sources (via Google search according to my interpretation of the feedback below).

  In the United States for example, there are many thousands of "best smartphone"-type searches each month on Google alone, according to data from Google Keyword Planner - we might assume that LLMs like Chatgpt might be used in a similar way as an “answer engine” to compare products etc.

   See the original response to my prompt “Best Smartphones 2025” below: (I removed the hyperlinks from the citations):

“Best smartphones of 2025 (as of Aug 17, 2025)

•  Apple iPhone 16 Pro Max — Best overall (iOS): class-leading camera and battery, long software support, Apple Intelligence features rolling out. Linked Citation: TechRadar
  

• OnePlus 13 — Best overall (Android): blazing performance, huge battery, great value among flagships; cameras good, if not best-in-class. Linked Citations: TechRadar & Tech Advisor
  

• Samsung Galaxy S25 Ultra — Best camera-forward Android: excellent all-rounder with strong battery life and AChatgpttools; top picks across multiple guides. Linked Citation: Tom's Guide
  

• Google Pixel 9a — Best under $500/£500: standout cameras and AI at a low price; 7-year updates. Linked Citation: Tom's Guide
  

• Samsung Galaxy Z Flip 7 — Best clamshell foldable: larger cover screen, slimmer design, refined AI features. Linked Citation: Tom's Guide
  

• Samsung Galaxy Z Fold 7 — Best book-style foldable: thinner/lighter, productivity champ; strong reviews across outlets. Linked Citation: Droid Life
  

• Battery life crown (not a single pick, but notable): OnePlus 13 and iPhone 16 Pro Max rank highly in recent endurance charts. Linked Citation: Tom's Guide”
  
  

My analysis (assisted by Chatgpt)

     Below based on using direct prompts I have attempted to show how the response to my prompt “best smartphones 2025” was created in Chatgpt 5. I used the direct feedback from the tool itself. The below shows how Chatgpt used external web searches to supplement it's pre-existing knowledge base while distinguishing between what information used in the response came from it's training in contrast with what came from up to date web sources.

    I chose this “best smartphones 2025” prompt as it demands “up to date” information: it is a prompt that (most likely) can not be resolved reliably from the information in the model's knowledge base alone

  The process of response formation Chatgpt stated was as follows:

      Chatgpt stated that to generate the above response it didn’t use just it's internal knowledge base (pre-training up to June 2024) it also used the retrieved corpus (web documents retrieved by web search) to generate the answer. The knowledge base provided context and background to interpret the retrieved facts, and the actual recommendations came from up-to-date sources.

   Here’s what Chatgpt stated that the pre-existing knowledge base of Chatgpt contributed to the above answer:

1. Model histories & brand reputations (pre-2025)

• Awareness that Apple’s Pro Max models (13 Pro Max, 14 Pro Max, 15 Pro Max) consistently topped “best iPhone” rankings in prior years.
  

• Awareness that Apple emphasizes battery life, cameras, and ecosystem longevity, (so when retrieval confirmed “iPhone 16 Pro Max is best overall,” Chatgpt stated it “knew” it was consistent with past trends).
  

• Awareness that Samsung’s Ultra models (S21 Ultra → S24 Ultra) historically were the company’s flagship with best camera and hardware. This background made the S25 Ultra’s placement unsurprising and aligned with retrieved claims.
  

  • Awareness that Pixel “a” series (Pixel 6a, 7a) were known as affordable phones with strong cameras, which matched with the retrieved Pixel 9a reviews.
    
    

2. Category framing knowledge

• The concept of “best overall,” “best budget,” “best foldable” is not invented by a single article but reflects how tech reviewers usually structure recommendations.
  

• Knowledge that foldables are a distinct product class, led by Samsung Galaxy Z Flip (clamshell) and Z Fold (book-style) lines since ~2020.
  
  

3. Long-term specs & product cycles

•  Knowing Apple releases new iPhones every September, Samsung launches Galaxy S in Q1, and foldables mid-year, helped ensure retrieved reviews were indeed current-generation devices.
  

• General knowledge of Snapdragon chip evolution (e.g., Snapdragon 8 Gen 1 → 8 Gen 3 → 8 Elite in 2025), which gave context when articles mentioned performance.
  
  

4. Evaluation

• Understanding what makes a phone “best” in consumer tech journalism: battery life, camera performance, AI/software support, update policy, price tiers.
  

• This guided how Chatgpt stated that synthesized retrieved claims into categories (“best Android flagship,” “best battery life,” etc.) instead of just listing phones without context.
  
  

5. Filtering & reliability judgments

•  Knowledge that Reddit or leaks are less authoritative compared to Tom’s Guide, TechRadar, or TechAdvisor. This Chatgpt stated helped to" discard noisy retrievals (like forums) even though they surfaced in the corpus".
  
  

    Chatgpt stated that, “the [above information from the] knowledge base contributed general, enduring facts about smartphone cycles, brand reputations, and [product] category structures. The retrieved corpus (RAG) contributed the specific, up-to-date 2025 evidence (which phones are currently ranked best, new models, battery test results).”

The RAG process used (end-to-end, factual)

Below is the actual process Chatgpt stated that it ran to contributed the specific, up-to-date 2025 evidence as part of the list above, describing each stage, the exact queries, how items were filtered / ranked, and how they fed into the final answer.

1) User query → intent

Input: “best smartphones 2025”.

Intent classification: comparative recommendation, up-to-date info required → browse web for recent buying guides, long-term reviews, and category lists (including foldables and budget).

2) Retrieval (web searches Chatgpt executed)

Chatgpt stated that it issued targeted searches to grab authoritative, current lists & reviews:

General “best phone” roundups:

• “Tom's Guide best phones” → Best phones 2025 tested. Source: https://www.tomsguide .com/best-picks/best-phones
  

• “TechRadar best phone 2025” → The best phone 2025. Source: https://www.techradar .com/news/best-phone
  

• “TechAdvisor best phone” → Best phones (Aug 15, 2025). Source: https://www.techadvisor .com/article/724318/best-phone.html
  
  
  [Product] "Category supplements":
  

• “Tom's Guide best battery life 2025” for endurance context. Source: https://www.tomsguide .com/us/smartphones-best-battery-life%2Creview-2857.html
  

• “Tom's Guide best foldable phones” for 2025 foldables. Source: https://www.tomsguide .com/best-picks/best-foldable-phones
  
  
  Specific “hero” models to validate status:
  

• “iPhone 16 Pro Max review TechRadar”. Source: https://www.techradar .com/phones/iphone/iphone-16-pro-max-review
  

• “OnePlus 13 review TechRadar / details”. Sources: TechRadar
  
  https://www.techradar .com/phones/oneplus-phones/oneplus-13-review
  
  https://www.techradar .com/phones/oneplus-phones/oneplus-13
  

• “Pixel 9a review Tom's Guide”. Source: https://www.tomsguide .com/phones/google-pixel-phones/google-pixel-9a-review
  

• “Z Fold 7 review” (to confirm 2025 generation strength). Source: https://www.droid-life .com/2025/07/24/galaxy-z-fold-7-review/
  
  

Why these? Chatgpt stated they’re "current, lab-tested roundups and deep reviews from established outlets, updated within weeks / days of today".

3.  Content processing
   

Chatgpt stated that it kept pages that met all of the following criteria:

• Published/updated in 2025 and recent (weeks to months). (e.g., TechAdvisor Aug 15, 2025; Tom’s Guide “best phones” updated last month). Sources: https://www.techadvisor .com/article/724318/best-phone.html & https://www.tomsguide .com/best-picks/best-phones
  
  

• Provide ranked lists or full reviews (not brief news or speculative leaks).
  

• Represent a mix of platforms / price tiers (e.g. iOS / Android, flagship / budget / foldable).
  
  Chatgpt stated that it "dropped" the following information sources:
  

• Forums/social posts and older listicles unless supporting color (e.g., Reddit threads), because they aren’t primary testing sources. (Searched but not used in synthesis.). Source: https://www.reddit .com/r/oneplus/comments/1hwf5lj/oneplus_13_review_philip_berne_is_dumbfounded/
  
  

4) Content/claim extraction

From each kept source Chatgpt stated that it pulled verifiable claims with dates. Examples:

• TechRadar ranks iPhone 16 Pro Max as “best overall phone.”. Source: https://www.techradar.com/news/best-phone
  

• TechRadar / TechAdvisor and Tom’s Guide position OnePlus 13 as a top Android pick, praising performance and battery. Sources:
  
  

https://www.techradar .com/phones/oneplus-phones/oneplus-13-review

https://www.techradar .com/best/best-android-phones

https://www.techadvisor .com/article/2562174/oneplus-13-review.html

https://www.tomsguide.com/best-picks/best-phones

•  Tom’s Guide highlights Galaxy S25 Ultra among best phones/best cameras. Source: https://www.tomsguide.com/best-picks/best-phones
  

• Tom’s Guide review: Pixel 9a is the best cheap phone around $499 with long updates. Source: https://www.tomsguide .com/phones/google-pixel-phones/google-pixel-9a-review
  

• Tom’s Guide foldables roundup places Z Flip 7 at the top; Z Fold 7 praised across long-form reviews. Sources: https://www.tomsguide .com/best-picks/best-foldable-phones https://www.droid-life .com/2025/07/24/galaxy-z-fold-7-review/
  

• Tom’s Guide battery leaderboard shows OnePlus 13 and iPhone 16 Pro Max near the top. Source: https://www.tomsguide .com/us/smartphones-best-battery-life%2Creview-2857.html
  
  

5.  Ranking & consolidation
   
   
   

Chatgpt ranked candidates by:

1. Cross-source consensus (appears in 2+ reputable guides).

2. Recency of the article/review.

3. Category coverage (overall, Android, iOS, budget, foldables, battery).

4. Testing rigor (lab tests, quantified battery/camera data).
   
   
   

     Chatgpt stated that this process “elevated” the following models: iPhone 16 Pro Max, OnePlus 13, Galaxy S25 Ultra, Pixel 9a, Z Flip 7 & Z Fold 7.

6) Information Synthesis (generation with guardrails)

• Chatgpt stated that it wrote “concise, category-based picks, only asserting what multiple sources back” (e.g., iPhone 16 Pro Max as best overall; OnePlus 13 as best Android; Pixel 9a as best budget). Sources:
  
  https://www.techradar .com/news/best-phone
  
  https://www.tomsguide .com/phones/google-pixel-phones/google-pixel-9a-review
  

• Chatgpt stated that it “avoided rumors and non-review forthcoming product information] leaks [as information sources}”.
  

• Chatgpt attached citations immediately after each claim.
  
  

    7. Verification loop

• Chatgpt stated that it “spot-checked each claim against at least one other current source” (e.g., Tom’s Guide vs TechRadar vs TechAdvisor).

Sources:

https://www.tomsguide .com/best-picks/best-phones

https://www.techradar .com/news/best-phone

https://www.techadvisor .com/article/724318/best-phone.html

https://www.tomsguide .com/us/best-phone-cameras%2Creview-2272.html

•      Ensured dates are in 2025 and recent (days–weeks) for foldables and camera/battery pieces. Source: https://www.tomsguide .com/best-picks/best-foldable-phones

8) Output formatting

Chatgpt stated that it:

• Presented a short ranked/category list for quick reading.

• Kept language neutral; flagged trade-offs (e.g., OnePlus 13 cameras “good, if not best-in-class”). Source: https://www.techadvisor .com/article/2562174/oneplus-13-review.html
  
  

    Chatgpt stated that (for clarity) in the construction of this response:

• No manufacturer press releases or “rumor blogs” were used to make picks.

• No sponsored shop links or affiliate pricing influenced rankings.

• No unverifiable specs were included.