How to get your app recommended by ChatGPT

When someone asks “what’s the best habit tracker for iPhone?”, a modern assistant usually doesn’t answer from memory alone. It runs a web search, reads a handful of pages and writes an answer from them. OpenAI describes its web search tool as giving models “up-to-date information from the internet” with “sourced citations”, and Gemini’s grounding with Google Search returns the web sources it used alongside the answer. Claude’s web search tool works the same way: Claude decides when to search, and the answer comes back with cited sources.

So two things decide whether your app gets named:

• What the model learned in training. Slow to change and out of your direct control.
• What the search step retrieves right now. Pages you can create, update and get yourself onto.

For recommendation questions, the retrieved pages are rarely your homepage. They tend to be listicles (“12 best habit trackers”), comparison posts, reviews, directories and community threads. The practical question becomes: do the pages that show up for your users’ questions name you, and describe you the way you want?

1. 01

   Write down the questions your users actually ask

   Aim for 10–30. Mix shapes: “best X app”, “X app for Y” (“budget app for couples”), “alternative to Z”, “free X app without ads”, “A vs B”. If you sell in several countries, note which questions matter where.

2. 02

   Ask them yourself, more than once

   Run each question in ChatGPT, Gemini and Claude at least three times. Log who gets named, in what order, and which pages are cited. The tracking guide has a template.

3. 03

   Say what you are in one sentence, in text

   Your site should state category, platform, audience and two or three differentiators in plain words near the top, not only inside screenshots. For example (a fictional app): “Margins is a journaling app for iPhone and Mac with end-to-end encrypted sync.” Use the same name and category everywhere.

4. 04

   Get onto the pages the assistants already cite

   Step 2 gives you a list of cited URLs. For each one, decide: a list whose author you can email with accurate facts, a directory you can submit to, a community thread where you can answer honestly (and disclose that you built the app), or a review site where you can ask real users for reviews.

5. 05

   Publish honest comparison and alternative pages

   “Alternative to Z” and “A vs B” questions are common. A fair comparison page on your own site, with a real table and the cases where the other app is the better choice, gives the search step something specific to retrieve.

6. 06

   Keep facts consistent and current

   Pricing, platforms, free tier and key features should match across your site, your App Store or Google Play listing, Product Hunt and review sites. Contradictions invite hedged or outdated answers.

7. 07

   Let the search crawlers in

   OpenAI says sites that opt out of OAI-SearchBot “will not be shown in ChatGPT search answers”. Gemini grounds on Google Search, so allow Googlebot, and leave Google-Extended allowed too: Google says it controls grounding in Gemini Apps. For Claude, allow Claude-SearchBot and Claude-User: Anthropic says blocking them may reduce your visibility in Claude’s search results. Details in how to get cited.

Mobile apps

Your App Store or Google Play page is a public web page and can be retrieved, but it’s a thin source: short text, few comparisons, and the App Store keywords field isn’t shown on it at all. Have a real website with a text description of the app, link it from your listing, and link the listing from the site. If nothing outside the store describes your app, the search step has very little to find.

Web apps and SaaS

For SaaS, questions are often about fit: “invoicing tool that integrates with Stripe”, “cheapest uptime monitor for a side project”. Public pricing, integration pages and docs give the search step concrete facts to match. Hiding pricing behind a demo form makes “cheap” and “free” questions hard to answer in your favor.

• Keyword stuffing. In the original GEO paper, keyword stuffing scored below unoptimized content on the main benchmark (17.8 vs 19.5), while adding quotations, statistics and cited sources scored highest.
• Special AI files as a shortcut. Google says you don’t need new machine-readable files or special schema to appear in its AI features. OpenAI’s crawler docs don’t mention llms.txt. Neither is a documented lever.
• Fake reviews and astroturfed threads. They break platform rules, get removed, and can surface in answers as complaints.
• Checking once. A single answer tells you little. Re-run the same questions weekly and watch the rate.

Nobody publishes a timeline, and anyone promising one is guessing. What we can say: changes to the pages the search step retrieves can show up once those pages are re-crawled, while what the model learned in training only moves with new model versions.

Set a baseline, make one or two changes at a time, and re-check weekly on the same question set so you can tell real movement from normal variance. The metrics guide shows how big a change needs to be before it means something, and MentionReach’s pay-per-check pricing keeps weekly re-checks cheap.