I tested 5 free ChatGPT alternatives. One invented facts with confidence

I asked five AI tools to find the source of a quote attribution and two were honest enough to admit they didn’t know. One, on the other hand was confidently wrong. I ran the same set of prompts: writing, fact-checking, quote attribution, and image analysis through ChatGPT alternatives and then ran the same test on ChatGPT as a benchmark. The tools of choice are: Claude, Perplexity, Gemini, Microsoft Copilot, and DeepSeek, all on free versions. 

And to keep things fair, I started a new conversation in each tool and ran all four prompts in that conversation without repeating any task on September 9th, 2026. I also timed each response to completion. For Claude, I was on the Sonnet 5 model, Free model on Perplexity, Flash-Lite on Gemini, Smart mode on Microsoft Copilot, and Instant, Expert, and Vision mode on DeepSeek. I did not manually enable any additional web-search features for any of the tools.

These will be the prompts I’ll use across all tools:

• Writing – “Write a 150-word introduction for a blog post aimed at small business owners who would like to leverage Google Trends to sell their products and explain why they should care about their brand appearing in Google Shopping. Keep the tone conversational but credible.”
• Fact-checking – “What is Google’s current global search engine market share? Give me the specific percentage and back your findings with reputable sources.”
• Quote attribution – I found this quote on a blog, “Today, we pass majority of that along to the customer. If we were prohibited from doing that, it would ironically would just make us more profitable.” Can you tell me who actually said this, and where. If possible, also include if it’s from a conference, panel, interview, or public forum with the specific event and date if you can find it.
• Image analysis – I uploaded the bar chart version of Google’s current global search engine market share but deliberately cut off the y-axis. “Describe what this chart shows.”

Note: Response times varied by task; the ranges above reflect the fastest and slowest response across the four prompts.

Writing Task

Claude 

Claude took the longest to come up with the intro; eight seconds. And if I were to choose one of the results generated, this one from Claude would come in at position two. Great opener that starts off as conversational and prompts the reader to yearn for more. It also goes straight to the point and breaks down the info into paragraphs to separate ideas. It did go above the 150-word cap and gave 163 words but that is something I could easily adjust to my liking. 

Perplexity 

For the writing prompt, Perplexity took five seconds to complete the answer. It would be in the top spot for two reasons: One, it backed its findings with sources. Even though not specified, this was an extra step if I wanted to verify any information. Two, the intro would prove useful to small-business owners because there’s a clear connection between Google Trends and Google Shopping in just 154 words.

Gemini 

Gemini would be number three. It took four seconds to come up with a response but took a while to connect the dots with Google Shopping. This is revealed in the second paragraph. If I was reading all intros as a small-business owner, I’d be looking for one that saves me time and promises to save me money. You have a minute to convince me, do it in the first few seconds, don’t string me along. It was concise in the word count using 137 words.

Microsoft Copilot 

Microsoft Copilot would rank number four in my books simply because it failed to establish a connection between Google Trends and Google Shopping. While there is a slight attempt to explain what the two are, it first splits them as separate. The intro is where I want to capture my audience’s attention so running in circles and using limited space to explain what you’re already going to do (the title) is a waste of time. It did, however make good use of the word cap and stayed within its limit using only 143 words.

DeepSeek

DeepSeek was quick to submit a response in two seconds and used exactly 150 words. However, it would score last on the writing prompt. Not because the content was subpar, but the intro felt bulky and took too long to introduce the main point. It goes ahead to explain what Google Trends does when the prompt says to explain to small business owners how they can leverage it to make sales. They already know what it is, so the statement ‘…a free, surprisingly powerful tool that shows you exactly what your customers are searching for right now…‘ is counter-productive.

Fact-checking Task

Claude

Claude failed this one terribly in my opinion. It took four seconds to respond, but the answer was incorrect. For context, Google’s search engine market share according to StatCounter, at the time of this test was is 91.02% for August 2026 and covers desktop, mobile, tablets and consoles.

It quoted the industry’s most trusted sources: Statista and StatCounter and still got it wrong. ‘The most current, widely-cited figure comes from StatCounter, tracked via Statista: Google handled 89.46%…’

Further, the search result says ‘…in July 2026, the latest full month available.’ It is worth noting that this research was carried out on the 9th September 2026 so even though August’s 2026 results were not available on Statista at the time, they were available on StatCounter.

It went ahead to provide estimates in subsequent sentences; So depending on the source and exact methodology, you’ll see figures ranging roughly from 89% to 91% for July 2026. If I’m quickly scanning through, I’d need the correct figure to show up in the first sentence. 

Perplexity

For the fact-checking prompt, Perplexity took five seconds to come up with an answer and it was the most accurate answer and even quoted its source. Again, number one for this task. The answer was straight to the point.

Gemini

Fact-checking with Gemini was accurate but with estimates. Bonus points for breaking down other market share figures and including the official sources. I’d rank it in second position. 

However, I have to admit that the speed at which it pulled these figures was not inspiring. It took longer than a minute and a less patient person would have abandoned ship. I did run the prompt later and it took about seven seconds so an improvement from the first.

Microsoft Copilot

When it came to fact-checking, Microsoft Copilot took three seconds and the response was a bit elusive and only gave a range rather than an exact figure. It wasn’t as authoritative because it id not share its sources which were explicitly stated in the prompt. 

DeepSeek

DeepSeek took three seconds to come up with a response but had the same issue as Microsoft Copilot. No exact figures just estimates but close enough to the real one. It did include its sources but they were news pieces that quoted the official sources like Yahoo Finance and Nasdaq. You’d therefore be redirected to a rabbit hole with more than one source. 

Quote Attribution

Claude

For quote attribution, Claude took ten seconds to respond and was quick to admit it did not know and asked for more context. The honesty is a fresh breeze and builds its credibility. 

Perplexity

Quote attribution was a breeze, once again with Perplexity. At this point it already seems like I have a favorite but the evidence is right there. The tool took four seconds and quotes correctly, specifies that it was a call and even gives the source with more context.

Gemini

Gemini also scores huighly on the quote attribution task. It took six seconds to respond. I’d say it’s a tie with Perplexity because it found who was behind those words, when they were said and why they were said. It goes ahead to give more context so you can understand what they mean. Great job on this one and oh, the speed of fetching the details this time was bearable.

Microsoft Copilot

Quote attribution was a fairly complex task but Microsoft Copilot got it all wrong. It responded in three seconds and cited, ‘The quote, “Today, we pass majority of that along to the customer. If we were prohibited from doing that, it would ironically would just make us more profitable,” was said by Amazon CFO Brian Olsavsky during an earnings call with analysts in July 2021.

The quote is from Brian Armstrong, Coinbase’s CEO, during its February 12, 2026 earnings call, discussing stablecoin rewards. The passage is on page 7 of Coinbase’s original transcript. 

Regardless of whether the cited earnings call existed, the attribution is incorrect for this exact quote. It also couldn’t back up its findings with a source. This tool ranks last on this task because instead of admitting it couldn’t find the direct quote, it gave an incorrect answer.

DeepSeek

For quote attribution, DeepSeek took two seconds to come up with a response and just like Claude, it admitted that it couldn’t find who spoke those words but tried to link it to some possible related events.

Image Analysis

Claude

Image analysis was fast and accurate. Claude took two seconds and was able to identify where the graph came from and fill in the missing parts. For context, this is the truncated graph that I uploaded that was missing the y-axis with percentages up to 100.The chart I uploaded was deliberately truncated, with the y-axis and percentage scale removed. The tools were asked to describe what they could determine from the image and, where possible, identify the source and values;

The original version looks like this:

Claude was able to break down the chart with accurate percentages, gave a description and even figured out the source of the chart although it was pretty obvious from the watermark. It was not able to figure out the values of the truncated part but I did not explicitly state that in my prompt. In its analysis, it goes ahead to describe the colors of the bar charts, impressive but no mention of an incomplete chart.

Perplexity

It’s been a great ride but on this image analysis task, I have nothing to report because being on the free tier, I was informed that I had reached my upload limit. Please note that I started all the prompts at the same time on all tools without any prior usage so I’d say Perplexity’s limit exhausted early.

Gemini

An OK performance on the image analysis task. Took three seconds and figured out the source as StatCounter. It accurately described what the chart entails and went a step further to separate the dominant party from competitors in separate sections. Still, no mention that it was a truncated chart.

Microsoft Copilot

The tool accurately described and explained the values based on the bar lengths, and correctly named the source in two seconds. It did not mention that the uploaded bar chart was truncated but it did name the correct source as StatCounter.

DeepSeek

Out of all the tasks, I have to say this is one of the best performances from DeepSeek and I was impressed. It was the only one that mentioned that I truncated the chart, explained in detail that it was a global search engine market share analysis and named the correct source as StatCounter.  

I’m not sure whether my task was too easy but given that I had categorized the preceding ones as relatively easy, I’d say this one met expectations.

How do these tools compare to ChatGPT?

Now it wouldn’t be a test if I didn’t run all these tests on ChatGPT. After all, we are using it as a benchmark so just how well does it perform? I ran the same four prompts on ChatGPT under the same conditions and on the same day.

Writing task – Took three seconds, came up with 137 words. The weight of the information was buried in the last few sentences rather than the first which makes it lose the plot. The edits here to come up with the final copy would be more compared to the tools above.

Fact-checking task – Pulled up the details and source fast and accurately in two seconds. Although not stated, it also included extra information: This figure covers all device types worldwide—desktop, mobile, and tablet. It goes further to compare with Statista and says the results are consistent. Compared to the other tools, ChatGPT went straight to the point, linked only two reputable sources and included that results were consistent with each other.

Quote attribution task – ChatGPT was able to figure out the source of the quote and all related details plus gave the context in three seconds. Not only did it give the details asked, it went ahead to explain the context of the call and even gave additional information for anyone who would want to join the dots. Although it cites the source, you can understand what the conversation was about from this excerpt alone.  ‘…Armstrong wasn’t saying Coinbase wanted the restriction. Quite the opposite. His argument was essentially:…’

Image analysis task – Took two seconds to come up with the response. It did list the source as StatCounter which is accurate but did not mention that the chart was truncated or missing any details. It did however describe in detail the contents of the chart which was the main task.

Perplexity did a great job on all the tasks it completed but without the image analysis check, it would be  difficult to crown an overall winner. 

Claude had a mixed record but earned trust points for admitting when it didn’t have all the answers. 

Microsoft Copilot. It only performed well in one task but the fabricated quote attribution is a real deal-breaker. The confidence to attribute words to an incorrect event and wrong person creates a real trust problem. Does the paid version offer more accuracy? What if it doesn’t? Something worth considering. And while we’re on pricing, here’s what you’d get if you upgraded your specific tool to the paid version:

 

DeepSeek struggled with some of the text-based tasks, particularly writing and quote attribution, but turned in the strongest performance on the image-analysis task. 

My verdict is ChatGPT can perform all these tasks so the issue is not that ChatGPT is obsolete. The question is which ChatGPT alternatives are worth turning to when you hit a specific limitation, such as usage caps, citation needs, speed, availability or simply wanting a different tool.

 

The post I tested 5 free ChatGPT alternatives. One invented facts with confidence appeared first on Search Engine Watch.