Gemini felt dumb on February 1, 2026.

I spent hours every day trying to get Gemini 3 Pro to obey my prompts, but it keeps forgetting instructions from just two messages ago. It keeps spitting out images and videos even when I explicitly say no, and it repeatedly uses forbidden words. The tool’s behavior is maddeningly inconsistent, especially since it used to work better last year. I’m desperate for settings or a workflow that will finally make it follow my directions.

I decided to test ChatGPT’s new ability to read a tiny screenshot of handwritten text—something Gemini could handle before but now seemed blocked by rate limits. I tossed the image at it, and the model parsed it without a hitch, proudly shouting “Bring it on!!”. The smooth, un‑filtered response felt like a win, making me confident that ChatGPT has finally caught up where other services fell short.

I tried using Gemini 3 Pro to follow a game guide, but the answers were completely off‑base, as if the model was hallucinating every step. Instead of clear directions, I got nonsense that made playing impossible and wasted my time. It felt dangerous and unreliable, especially compared to Grok, which actually got the guide right.

I tried Gemini’s multi‑step agent with a vague prompt about a young woman in a Ferrari jacket, and it automatically refined the description into a hyper‑detailed, photorealistic scene—perfect pose, night lighting, sponsor logos, and crowd blend—all in one go. The result was insanely coherent and detailed, making the tool feel like a creative partner that exceeded my expectations.

I tried the “Cold Boot” prompt with Gemini 3 Pro after a three‑month break, uploading my whole repo. The AI acted like a tireless project manager, telling me exactly where I’d left off and the next tiny step. When it pointed out a missing bracket, I fixed it and instantly felt back in the flow. The whole restoration took seconds and erased the inertia that usually stalls me.

I tried the Gemini CLI hoping it would match AI Studio, but it fell short. Forgetting to run the offline save_memory command wiped whole sessions, and the pro tier capped prompts before forcing a switch to Flash. The hard‑coded system prompt kept it boring and code‑centric, never feeling like Studio. To top it off, a persistent looping bug sometimes spewed endless nonsense, making the experience irritating and unreliable.

I tried three times to get Gemini to check a single sentence’s grammar, but each response was the same annoying placeholder text instead of any useful correction. The repeated irrelevant output made the whole interaction feel pointless and frustrating, leaving me without the help I needed.

I’ve been using Gemini CLI a lot lately and, honestly, it’s a mixed bag. The tool constantly asks for explicit prompts to stay on track, which is annoying, but when it does focus it churns out surprisingly solid UI/UX suggestions that follow best practices. I normally lean on Codex, yet Gemini feels like it could outrun it—especially with its knack for crafting Reddit‑style comments that are almost indistinguishable from a real human. The potential is there, it just needs more polish and attention.

I tried using Google’s AI and kept noticing it peppered every sentence with extra commas. It wasn’t breaking anything, but the over‑use of punctuation made the responses feel stilted and hard to read. I found myself pausing to mentally edit the output, which was mildly irritating and left the overall experience feeling just average rather than impressive.

I spent time chatting with Gemini 3 Pro on the web app and quickly got fed up. The model kept “hallucinating” – it would subtly change the meaning or name of concepts mid‑conversation, leaving me confused about what it was actually answering. Even short back‑and‑forth felt unreliable, and the constant term‑shifts made the whole discussion feel pointless.

I’ve been a Pro Gemini subscriber for six months, and only recently discovered that the real “effective” context window is just 32k tokens—far less than the advertised 1 million. When I upload a document, the model skims most of it, and AI Studio cuts me off after a single prompt. The free tier’s low limits make sense, but the paid plan’s restrictions feel like a joke. I’m frustrated that the premium version can’t even handle the promised capacity.

I tried using the AI tool and kept hitting the same error that’s been around since 2026. The repeated failure messed up my workflow and left me feeling stuck and frustrated, because the problem still isn’t resolved and it feels like a huge waste of time.