Ask ChatGPT whether it would recommend a SaaS tool. Then ask Perplexity the same question. You will probably get two different answers, and at Inithouse, we can measure exactly how different they are.
We built Be Recommended to run that test at scale. The tool queries five AI engines: ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews. It sends the same prompt about the same brand to all five, then scores how strongly each engine recommends it on a 0–100 scale. After running thousands of these checks across our own portfolio and for early users, one pattern kept showing up: the same brand rarely gets the same score twice.
The gap is real and measurable
We pulled data from recent runs across multiple brands and product categories. A brand might score 72 in ChatGPT and 31 in Perplexity. Same query, same day, same product. The average cross-engine spread (highest score minus lowest) sits around 25–40 points on the 0–100 scale. That is not noise. That is five different recommendation systems reading the same internet and arriving at different conclusions about who to suggest.
The biggest gaps tend to appear between ChatGPT and Perplexity. Claude and Gemini usually land somewhere in the middle. Google AI Overviews behave differently again. They pull from search index data, so they correlate more with traditional SEO signals than the others do.
Each engine has a source preference
When we dug into why the scores diverge, a pattern emerged in what each engine treats as a trustworthy signal.
| Engine | Preferred source type | What moves the score |
|---|---|---|
| ChatGPT | Authoritative, editorial content: established publications, well-structured documentation, Wikipedia-adjacent references | Brands with press coverage, detailed product pages, and third-party reviews score higher |
| Perplexity | Community and forum sources: Reddit threads, Hacker News, Indie Hackers, Stack Overflow, developer forums | Brands that people actually discuss in public score higher, regardless of editorial coverage |
| Claude | Structured product documentation, technical specs, comparison pages | Clear, factual product descriptions and well-organized docs tend to rank well |
| Gemini | A mix. Pulls from both editorial and community, but weights Google's own index heavily | Brands ranking well in traditional search tend to carry over, but grounding citations can be inconsistent |
| Google AI Overviews | Search-index first: existing rankings, snippet-eligible pages, schema markup | Closely tied to organic SEO performance; a page ranking on page 1 for a query is more likely to be cited |
This table is a simplification, and the weighting shifts with every model update. But the directional pattern has held across months of our measurements: ChatGPT leans editorial, Perplexity leans community, Google AI Overviews lean search.
What this means in practice
If your content strategy only targets one source type (say, publishing blog posts optimized for Google), you are building visibility for at most two of the five engines (Google AI Overviews and partially Gemini). You are invisible to the other three until someone writes about you on Reddit or a reviewer picks you up.
We see this in our own portfolio. One of our products had strong Perplexity scores because users were discussing it on Indie Hackers and Dev.to, while its ChatGPT score stayed low because no established publication had covered it. Another product had the opposite pattern: good documentation and a few press mentions pushed its ChatGPT score up, but zero community discussion kept its Perplexity score near the bottom.
The practical takeaway is unglamorous: you need presence across source types. Editorial mentions, community discussions, structured documentation, comparison pages, forum threads. Not because any single one of these guarantees a recommendation, but because each engine has its own idea of what counts as evidence that a brand is worth suggesting.
The measurement itself is the product
Be Recommended exists because we needed this data for our own products. We run a portfolio of tools at Inithouse, and tracking how each AI engine perceives each product turned out to be the most actionable feedback loop we had. Watching a score shift after a specific action (a new blog post, a Reddit thread, a product update) tells us more about what worked than any traffic dashboard.
The 0–100 score per engine is the simplest version of it. The report also breaks down which queries trigger a recommendation and which do not, so you can see exactly where each engine draws the line for your brand.
If you are shipping a product and wondering whether AI assistants are sending people your way or to a competitor, the five-engine comparison is where we would start. One engine's opinion is anecdotal. Five engines disagreeing about you is data you can act on.
We build at Inithouse. A studio shipping a growing portfolio of AI-powered products. Be Recommended started as an internal tool and became one of the most used products in the portfolio. You can run your first check at berecommended.com.
Top comments (0)