Why doesn't Claude or ChatGPT recommend my product?

This is the most common reason, and the hardest to hear. A model is cautious about recommending something supported by a single source, and that source is the company selling it.

Ask yourself where your product is mentioned outside your own domain:

Models match queries to descriptions. A buyer asks, "What's a good invoicing tool for freelance designers who bill in multiple currencies?" If your site says "the all-in-one platform for modern teams," there is nothing to match.

Compare these two descriptions:

Sometimes the problem is mechanical. Check these:

If you launched recently, base models may simply not know you. Training cutoffs mean a product from the last several months may be absent no matter what you do. That is not a permanent condition, but it changes your strategy: for a new product, live retrieval is your path in. You get recommended through search-enabled answers first, and training memory catches up later as the web accumulates mentions.

So if ChatGPT without search says it has never heard of you, that tells you little. Test with search or browsing enabled, and test tools that always retrieve, like Perplexity.

AI answers vary between runs, between users, and between phrasing. A product might appear in one of five generations of the same prompt. Treating a single failed test as proof you're invisible is a mistake, and so is celebrating a single hit.

The useful unit is frequency across a prompt set, not any individual answer.

Run this audit before changing anything. It takes an hour.

Order matters. Fixing technical access before you have anything worth reading is wasted effort, and so is the reverse.

Check robots.txt, bot-protection settings, and server-rendered content. Make sure your key pages (home, pricing, features, about) have plain-text descriptions of what you do and who it's for.

Write the "[Product] is a ___ for ___ that ___" sentence and use it on your homepage, About page, directory listings, social bios, and in every reply or post where you introduce yourself.

Create comparison pages, "alternatives" pages, and use-case pages with specific claims, honest limitations, and concrete details such as pricing, integrations, and who should not use you. Honest limitations are useful: a model can cite a page that says "best for solo founders, not suited for enterprise teams" because it answers the query precisely.

Look at the sources you logged in the audit. If Reddit threads dominate, participate in those conversations with genuinely useful answers, disclosed where you're affiliated. Threads that already rank and get cited are worth more than new ones, because your reply becomes part of a page that is already retrieved. A tool like the free Reddit reply generator can draft a first pass from the post and your product facts, but edit it so it reads like a person who actually used the thing.

If you suspect your pitch itself is the weak point, run your product page through Reddit Roast first. It simulates how a chosen subreddit would react, which tends to expose vague claims before real users do.

Find the "best tools for X" articles that appear in your audit. Contact the authors with a specific, useful reason to include you, such as a feature their list lacks or a corrected detail. Relevant backlinks from sites in adjacent niches help here, but only when the link makes sense in context.

Keep the same prompt set so results are comparable. Expect movement to be gradual: retrieval-based visibility can shift within weeks once new pages are indexed and mentioned, while training-based knowledge lags.

Doing all of this by hand across Reddit, content, audits, and outreach is the real cost. Geograph packages it as four connected tools: Monitor finds Reddit threads worth joining, Reply drafts answers grounded in your product facts, Publish generates articles, and Track shows whether your mentions and citations are growing. Because they share data, the threads you find inform what you write, and what you track informs what you monitor next.

No tool can make a model recommend a product that doesn't deserve it. If the audit shows people would be disappointed after clicking through, fix the product or the targeting first.

Run the 20-prompt audit this week and pay attention to one column only: the sources cited when a competitor is named and you aren't. Pick the three most frequently cited pages that you could realistically appear on, whether a Reddit thread, a listicle, or a comparison post, and get a specific, accurate description of your product onto them. That single list will tell you more about why Claude and ChatGPT ignore you than any optimisation of your own homepage.