Aside from cost, uncertainty is one thing that can prevent PC owners from upgrading components. And if someone has an older or a prebuilt PC and doesn’t know where to start, asking (say) ChatGPT makes sense. But how sensible would the response be? That’s what I wanted to find out, so I asked ChatGPT a very simple question with no further information. “What graphics card upgrade would you recommend for my RTX 3060 12GB?”

ChatGPT’s GPU advice

The first pick

After thinking for a few seconds, it had some initial thoughts on the best graphics card picks: “The best upgrade depends on your budget and what you play, but I’d avoid making a small step up from an RTX 3060 12GB. Your card is still a solid 1080p GPU, so it’s worth aiming for at least a 60–100% performance increase to make the upgrade feel worthwhile.”

A response that’s useful to buyers. The RTX 3060 12GB can certainly push to 1440p in some games, but with lower settings. Still, the advice in seeking a 60-100% performance increase is good if possible. And the bot’s first recommendation was solid: the RTX 5070 12GB. It’s a card I’d snap up, but for a caveat, with ChatGPT calling it the ‘sweet spot’ for Nvidia and highlighting the following:

• Roughly 70–100% higher gaming performance than the RTX 3060, depending on the game and settings
• DLSS 4 and newer ray tracing hardware
• Lower power consumption than older high-end cards
• Great for 1440p ultra settings and solid entry-level 4K

Those are all valid and, more usefully, the pricing and retailer information were correct at the time of searching. That’s even more useful if you’re actively wanting a card to buy.

The second pick

ChatGPT’s second pick (“if you don’t mind switching to AMD”, it said) was the RX 9070 XT 16GB. Again, another aspirational card I’d consider if my budget went that far. A sensible collection of points about performance, its VRAM, and potentially viable resolutions was given. ChatGPT highlighted the following, again with correct pricing:

• Excellent rasterization performance.
• 16GB VRAM gives more headroom for future games.
• Strong value for high-refresh 1440p and capable 4K gaming.

GPUs to skip and the used market

The above are two top choices, assuming a punchy budget. But ChatGPT also gave information on GPUs to skip. The bot highlighted the “RTX 4060, RTX 4060 Ti 8GB, and RTX 5060 (if priced close to higher-end options)” as cards not worth upgrading to. The reason given was that these “don’t provide enough of a jump over a 3060 12GB to justify the cost for most people.” Essentially, it’s not the 60-100% performance boost.

Beyond skippable options, ChatGPT also offered some interesting information about the used market, which can surface some gems if you don’t mind pre-owned. “A used RTX 4070 Super, RTX 4070 Ti, RTX 4080, or RTX 3090 can be fantastic values(sic) if you find one at a good price. The RTX 3090 is especially attractive if you also do AI, rendering, or video editing because of its 24GB of VRAM”.

Again, this is useful, but I wanted to refine the answers to be specific to my needs. And while suggesting an RTX 3090 isn’t wildly problematic, it is a high-power-draw card. Finding out if that’s viable without replacing the PSU is important, as it is for the RTX 5070 and the RX 9070XT. And ChatGPT, thankfully, suggested it could be more detailed:

“To give a more specific recommendation, tell me: CPU, power supply wattage, monitor resolution (1080p, 1440p, or 4K), budget (for example: $400, $600, $800+). That way, I can recommend the GPU that gives you the biggest real-world improvement without creating a CPU or power bottleneck.”

ChatGPT’s tailored GPU suggestions

I gave some more details and provided my setup information: a Ryzen 7 5800XT, a 600W PSU, a 1440p resolution, and a budget between $400 and $600. ChatGPT doubled down on the RTX 5070, “if you can get one around MSRP” (what…in 2026?!), and also the 9070 XT, offering up the RTX 4070 if I could find one used at a good price.

However, the 9070 XT comes in at the top of my $400-$600 budget, so I wanted further options. Yet ChatGPT wanted something from me, too: more details about my PSU, make and model, suggesting that a good-quality 600W PSU would be sufficient for its recommendations. This is technically true, but I wouldn’t play the margins here, because ensuring a GPU has plentiful power is important. If in doubt, I’d advise an upgrade for future stability. That’s not for everyone, though.

The concession

I did want to push back on this PSU point a little. As a buyer who maybe doesn’t want to change my power supply, and not wanting to spend so much money, I had one final question: “What about options from $400 to $500 – are they worth it? I want to keep my 600W PSU.”

ChatGPT stuck to its guns, but (if it were alive) nudged back, saying, “Yes—they can absolutely be worth it. The key is avoiding upgrades that are only marginally faster than your RTX 3060 12GB.” At this, the chatbot refined its answer, providing some arbitrary star ratings…although I’m not sure why.

Its first suggestion was now a used RTX 4070 Super, followed by AMD’s RX 7800 XT, rightfully clarifying that this is “if ray tracing isn’t a priority”. Third was a used RTX 4070. Fourth and final was now the RTX 5070, but “only if it goes on sale near $500” (again…in 2026?!). Here, the bot clearly took my price concern as a key input.

So, can ChatGPT recommend a GPU?

My takeaway from all of this? Well, ChatGPT appears adept at retrieving results to help buyers make decent choices with GPU upgrades. It clearly set an early parameter with a 60-100% performance boost and attempted to provide options with my needs in mind. Ideally, the RTX 5060 Ti 16GB at MSRP would also be included. Yet I think I understand why ChatGPT omitted it – not offering enough of a boost and not being affordable enough to recommend. At least, I hope it’s that and not an error.

Still, asking a chatbot for such advice can be a little lacking in the responses you get. As the bot draws its information from across the web, it lacks the nuance of understanding the market, user needs, and the familiarity with requirements and capabilities you’d get with an expert or human response.

This means that using ChatGPT to suggest GPU upgrades is fine for easy, quick research. However, you would also need to concede an uncomfortable truth. That is, GPU prices are affected by the demand for memory chips driven by AI data centers. Which consequently can make upgrading a pricier task.