Leaderboard · B2B software
AI startups
Marketing plays for ai startups, ranked by votes.
Photo: Jeremy Waterhouse
-
1
Send a weekly email showing one real model output from that week
Each week, pick one real, representative output your product generated for an actual use case, with identifying details removed or replaced. Send a short email to your list showing the input prompt or trigger and the resulting output side by side, with two sentences on what changed since last week's version.
-
2
Post a LinkedIn writeup of one hallucination and how you fixed it
Document a real case where your model produced an incorrect or fabricated output, what caused it, and the specific evaluation or guardrail change that addressed it. Post it on LinkedIn with the before and after example, written plainly without minimizing the original mistake.
-
3
Partner with a data provider on a bundled evaluation dataset
Identify a data or evaluation provider relevant to your product category. Co-build a small benchmark or evaluation dataset relevant to your use case, publish the methodology and results jointly, and offer prospects a way to run the same evaluation against their own data during a trial.
-
4
Host a technical webinar walking through your evaluation harness
Run a 45 minute webinar where an engineer walks through the actual evaluation harness used to test model changes before release: the test set, the scoring method, and one recent example of a change that failed evaluation and was not shipped. Open it to technical evaluators, not a general marketing list.
-
5
Publish a written explanation of exactly what data trains your model
Write a plain language page explaining specifically what customer data, if any, is used for model training or fine-tuning, what is excluded by default, and how a customer can opt out. Link it from your privacy policy, pricing page, and every sales deck slide that mentions the model.
-
6
Request reviews on Product Hunt tied to one completed workflow
After a customer completes a full workflow using your product for the first time, such as generating a finished draft they actually used, send a direct request for a review on Product Hunt or a public software review site that asks them to describe that specific workflow rather than a general opinion.
-
7
Run the model on the customer's own input during onboarding
Design the first onboarding screen so the new signup pastes or uploads their own real content and sees the model's output on it immediately, before any tour or feature walkthrough. Follow with a short email showing how to refine that same output using a feature they have not tried yet.
-
8
Start a community channel for prompt and workflow sharing
Set up an invite-only community channel for paying customers to share prompts, workflows, and configurations that work well for their use case. Seed it weekly with one example from your own team and highlight one member contribution in your product newsletter.
-
9
Give sales an answer sheet for the model provider and version
Write a one-page internal reference for sales listing exactly which underlying model or models power the product, the version currently in production, and how model updates are tested and rolled out. Require sales to reference this directly rather than answering from memory, and update it the same day any underlying model changes.
-
10
Offer a fixed scope pilot with a success measure set upfront
Offer prospects a two-week pilot scoped to one specific task and one specific, jointly agreed success measure defined before the pilot starts, such as completing a set number of real documents to an accepted quality bar. Document the agreed measure in writing before granting access.
No plays yet for AI startups. Submit the first.