In the Weights turns AI model recall into the new vanity search
A new tool from Thomas Dimson and Joey Flynn scores how well major AI models remember you—without peeking at the web.
What matters
- In the Weights, created by Thomas Dimson and Joey Flynn, scores how well AI models recall a person from training data without web search.
- The site queries models including Grok, Gemini, multiple GPT versions, Claude, and Llama, then clusters responses and assigns a strength score.
- TechCrunch's Anthony Ha scored 641, placing in the top 6% of tracked names.
- The tool reframes vanity search for the AI era, where chatbot recall may matter more than search-engine indexing.
- Methodology details—such as how retrieval is prevented and how scores are normalized across models—remain unclear.
What happened
On June 20, 2026, TechCrunch's Anthony Ha reported on a new website called In the Weights, created by Thomas Dimson and Joey Flynn. The site asks a simple question: when an AI model is prompted to recall who you are—without using web search or any external tools—how well does it do?
The "weights" in the name refer to the numerical parameters that shape a model's training and output. As the site puts it: "Being in the weights means your existence was deemed important in the process of creating superhuman artificial intelligence."
Here's how it works, according to the report: In the Weights queries multiple models—including Grok, Gemini, several versions of GPT, Claude, Llama, and lesser-known models—with a prompt along the lines of "Who is [name]? Give up to 10 results, each with a short description and confidence." It then clusters similar descriptions together and assigns a strength score.
Ha himself received a score of 641, placing him in the top 6% of names tracked by the site—though he noted that several TechCrunch colleagues scored even higher, and that the leaderboard was shifting in real time as he wrote.
Why it matters
For years, "Googling yourself" was the informal benchmark for digital presence. If you showed up in search results, you existed in the public record. In the Weights reframes that question for the AI era: it's not whether a search engine can find you, but whether a language model already knows you from its training data.
This distinction matters because chatbots are increasingly where people get information. If a model can recall your name, role, and accomplishments without browsing the web, that means you were part of the corpus deemed significant enough to bake into the model's parameters. In the Weights turns that latent visibility into a quantifiable, comparable metric.
The tool is partly a vanity exercise—Ha openly frames it as the "new AI-centric vanity search"—but it also surfaces a real shift in how reputation and discoverability work. As more users rely on AI assistants for answers, being "in the weights" may become a more meaningful signal of cultural or professional footprint than traditional search ranking.
There are open questions about methodology. The site doesn't fully detail how it prevents models from using retrieval or browsing features, how it normalizes scores across models with different knowledge cutoffs, or how the clustering algorithm handles ambiguous names. The leaderboard's real-time shifting also suggests scores are sensitive to model updates or prompt variation.
Public reaction
No strong public signal was available from Reddit or other discussion forums at the time of writing. The story is fresh, and community reaction has not yet surfaced in the captured sources.
What to watch
- Whether In the Weights publishes its methodology in more detail, particularly around how it ensures models aren't using web search or retrieval-augmented generation.
- How scores change as models are updated or retrained—whether "being in the weights" is stable or ephemeral.
- Whether this concept evolves from a novelty into a more serious reputation or brand-monitoring tool, especially for public figures and companies.
- How model providers respond to being benchmarked this way—whether they see recall-based identity scoring as a meaningful evaluation axis.
Sources
Public reaction
No Reddit or public discussion data was available at the time of writing, so community sentiment could not be assessed.
Open questions
- Will users treat strength scores as a serious reputation metric or purely as entertainment?
- How will the community scrutinize the tool's methodology around preventing model web access?
What to do next
Developers
Try querying your own name across multiple models manually to compare recall quality, then check your In the Weights score for discrepancies.
Understanding how different models encode your identity in their weights can inform how you think about model training data and personal visibility.
Founders
Check your company name and key team members on In the Weights to gauge whether AI models recognize your brand from training data alone.
As chatbots become a primary information channel, brand recall inside model weights is an emerging visibility metric worth tracking.
PMs
Evaluate whether your product or feature names appear in model recall and consider how AI assistants describe your product unprompted.
If users ask AI assistants about your product category, the model's built-in knowledge shapes first impressions before any web result is returned.
Investors
Monitor whether In the Weights or similar tools evolve into reputation-analytics platforms that brands and individuals pay to track.
The shift from search-based visibility to weight-based visibility could create a new category of monitoring and optimization tools.
Operators
Search for your executive team and organization on In the Weights and note how models describe you, flagging inaccuracies or gaps.
Model-generated descriptions are increasingly what users see first; knowing what models say about your organization helps you correct or reinforce that narrative.
How to test
- 1Navigate to the In the Weights website.
- 2Enter a name into the search field and submit the query.
- 3Review the returned strength score, percentile ranking, and the clustered descriptions from each model.
- 4Compare results for multiple names (e.g., yourself vs. a well-known public figure) to contextualize the scoring scale.
- 5Re-run the same query later to check whether scores shift over time, as the TechCrunch report noted leaderboard movement.
Caveats
- The site's methodology for preventing models from using web search or retrieval is not fully documented.
- Scores may vary due to model updates, prompt phrasing, or clustering algorithm changes.
- Ambiguous or common names may produce conflated or inaccurate results.
- The tool is new and its scoring scale is not yet independently validated.