Why doesn’t ChatGPT mention our newly launched product?

Why can’t ChatGPT recommend a product it has never seen?

An AI assistant answering from memory only knows what was in its training data. Sharma (opens in a new tab), a single-author study from IIT Patna, tested 112 startups from the 2025 Product Hunt leaderboard on the developer version of ChatGPT (gpt-4o-mini) with no web search. He describes a “recency wall”: products launched in January 2025 cannot appear in that model’s training data.

He calls this “a hard constraint, not something optimization can overcome”. No change to your website can teach a model about a product it was never trained on. Only a newer model, or a live web search, can.

A second effect compounds the first. Even before the cutoff, established products have far more written about them, so there is more for a model to learn from. Sharma calls this “authority concentration”, and it means a new entrant starts behind even once it is old enough to be included.

Does ChatGPT search the web for new products?

Often, but not always, and that decides whether a new product can appear at all. In our hidden searches study, the consumer ChatGPT app ran a mean of 3.7 web searches per buyer question. None of the 42 answers without a search, across the assistants we tested, cited anything.

The same assistant can answer either way. In Kumar’s (opens in a new tab) tracking setup at the AI visibility company Ranqo, ChatGPT and Claude ran web search only on a weekly cycle per brand and answered from memory in between. Whether your launch can show up therefore depends on whether a search happens, and on what it finds.

How much do search-connected assistants help a new product?

They help, but less than many founders hope. In Sharma’s test, Perplexity, which searches the web on every question, surfaced the startups in 8.29% of discovery answers. That is 2.5 times ChatGPT’s rate, and still low.

Sharma puts the ChatGPT figure bluntly: a 3% discovery rate means 97 out of 100 relevant questions will not mention your product. For ChatGPT, nothing he measured predicted which products got through. For Perplexity, links from other websites, a strong Product Hunt ranking and genuine Reddit discussion all went with more discovery.

Does an optimized launch page get you found?

Not on its own, in the one study that tested it. Sharma scored each startup’s website for the features usually recommended for AI visibility, such as statistics, citations and structured data. That score showed no link with discovery on either assistant.

His reading is that on-page polish works as a multiplier, not a starting point: “You can’t multiply zero.” A product first has to appear on pages that the assistant’s searches find.

Those pages are often not the ones ranking on Google. In our study of AI citations and Google rankings, only 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the question asked.

Does fresh content help a new product get cited?

In assistants that search, recent pages have a measurable edge. In our freshness study, the pages ChatGPT, Gemini, Perplexity and Claude cited were first published about half as long ago as Google’s top 10 for the same questions, a ratio of 0.50. Between 17.4% and 22.6% of each assistant’s dated citations were under 90 days old, against 6.9% for Google.

A lab test points the same way. Researchers at the software company Sprinklr (opens in a new tab) ran 252,000 head-to-head trials on six AI models, each comparing two otherwise identical pages. A recent date was one of four factors that decided which page was cited first on all six models. The test compared content dated 2026 with the same content dated 2019.

Do not mistake a new date for new content. In our data, 66.4% of the recently dated pages in Google’s top 10 were old pages with a new modified date. The assistants’ edge came mostly from pages that were genuinely new.

How long until AI assistants know about a new product?

No study has measured that for brand-new products yet. Kumar reports that established brands were recognized right away when a question named them, from 94% to 100% across engines, while unbranded questions surfaced them far less often, at 12% to 52%. He cautions that a brand with no prior web presence can take longer.

For assistants that answer from memory, Sharma’s advice is stark: “there’s not much to do except wait” for newer training data. That timeline is set by the AI companies, not by you.

What should you do about it?

Plan a launch that does not depend on early AI discovery, while building what search-connected assistants can find.

1. Assume early AI visibility will be low. Budget for channels you control, and treat any AI mentions as a bonus at first.
2. Get written about on other sites. Reviews, launch platforms, comparisons and genuine community discussion went with discovery on the assistant that searched.
3. Publish clear, dated, factual pages. Put prices, specifications and launch dates on pages that comparison writers and assistants can quote. Device makers face this with every launch, since assistants must get launch and compatibility facts right, as how consumer tech brands reach AI shortlists shows.
4. Test the assistants your buyers use. Check whether they search for your category, and ask the same questions several times.
5. Track monthly. Visibility can change as new pages are found and new models are released.

If you want help planning launch visibility, see our approach to generative engine optimization.

What does the research not tell us yet?

The evidence on new products is thin, and most of it comes from one study.

• One small study carries most of the weight. Sharma tested 112 startups, skewed toward developer, productivity and AI tools, on two developer versions of assistants in December 2025.
• The measure is simple. A success counted whenever the product name appeared in the answer.
• Nobody has tracked time to visibility. How long a new product takes to appear in each assistant is still unmeasured.
• Freshness findings are associations or lab results. Neither our data nor the Sprinklr test shows what happens when a real launch page is published.

Frequently asked questions

Why doesn’t ChatGPT know about my new product?

If it answers from training data, your product may simply postdate it. In one study, ChatGPT without web search surfaced only 6 of 112 recent startups in any discovery answer.

Does ChatGPT search the internet before answering?

Often. In our study, the consumer ChatGPT app ran 3.7 web searches per buyer question, but answers that ran no search cited no pages at all.

Does Perplexity find new products better than ChatGPT?

In one test, yes: Perplexity surfaced new startups in 8.29% of discovery answers against 3.32% for ChatGPT without search, about 2.5 times as often.

How long does it take for a new product to appear in AI answers?

No study has measured it yet. Assistants that search can find new pages quickly, while answers from training data wait for the next model update.

Sources

• Sharma (2026), The Discovery Gap: How Product Hunt Startups Vanish in LLM Organic Discovery Queries (opens in a new tab), arXiv:2601.00912.
• Kumar (2026), Generative Engine Optimization at Scale: Measuring Brand Visibility Across AI Search Engines (opens in a new tab), arXiv:2606.20065.
• Vishwakarma, Kumar and Jamidar (2026), What Gets Cited: Competitive GEO in AI Answer Engines (opens in a new tab), arXiv:2605.25517.
• Underneath (2026), The hidden searches AI assistants run before they answer
• Underneath (2026), How fresh are the pages AI engines cite?
• Underneath (2026), Do ChatGPT, Gemini, Perplexity and Claude cite pages that rank?