ChatGPT Recommendation Optimization: A Practical Guide
Buyer-shaped prompts, memory versus browsing, comparison pages that get quoted — and how to measure the before and after.
Recommendation answers are where AEO stops being abstract: a buyer asks ChatGPT what to use, the answer names three products, and your pipeline either grows or does not. You cannot buy a slot in that list and you cannot prompt your way into someone else's chat. What you can shape is what the model knows about you, what its retrieval finds, and how consistently the two agree. The measurement loop this guide depends on is in How to Track Your Brand Mentions in ChatGPT, Gemini and Perplexity; start there if you have no baseline.
Start with buyer-shaped prompts
Buyers rarely type the keyword-shaped queries SEO taught us to chase. They describe their situation: "best CRM for a two-person agency", "tool to keep client projects on track without a project manager", "X or Y for a small ecommerce brand". Recommendation work begins by collecting that language — from sales calls, support tickets, community threads — and freezing ten to twenty prompts per category, comparison and problem shape. The set is your instrument; if it drifts, so do your numbers.
What the model knows versus what it retrieves
ChatGPT answers from two places. Training knowledge is what the model already believes about your category and your brand — built from the public web as it existed when the model was trained, and remarkably resistant to quick fixes. Browsing and retrieval add what it finds today: your current pricing page, fresh reviews, documentation. A recommendation often blends both — the shortlist from memory, the specifics from search.
The mix tells you where to push. Consistent, specific, widely corroborated facts feed the memory side over time; retrievable, well-structured pages feed the browsing side immediately. Neither can be hacked in a weekend, which is the honest reason to start now. The mechanics behind retrieval are in How AI Answers Pick Their Sources.
The consistent-facts layer
Models hedge away brands they cannot pin down. One name everywhere (not the product name on the site, an abbreviation in directories, and an old name in a press archive), one category, one sentence describing who it is for, and a pricing story that matches across pages. Contradictions typically live in forgotten places: legacy pricing pages that still resolve, renamed tiers, old directory listings, boilerplate in partner sites. Reconciling them is unglamorous and disproportionately effective — the audit that finds them is in The AI Answer Audit Checklist.
Comparison pages that get quoted
When ChatGPT browses for a recommendation, it leans on comparison material — and most vendor comparison pages are unreadable to a model because they compare against strawmen. The pages that get lifted take real positions: a criteria table with named use cases, a "who this is for / who it is not for" section, specifics instead of adjectives, and honest trade-offs. Honesty is not just ethics here; a comparison page that contradicts every third-party source gets discounted, while one that agrees with them gets quoted.
The third-party footprint
Much of what the model believes about you is written on sites you do not control: reviews, roundups, documentation written by your integrators, Reddit threads you did not start. You cannot edit those, but you can influence them the old way — give reviewers, partners and directory editors accurate, current facts, and make your own docs quotable so third parties have something true to copy. Corroboration across those sources is what turns a claim into a belief.
Measure the before and after
Run the frozen prompt set before you change anything, then monthly after. Expect gradual movement, not a switch: months, not weeks, is typical for recommendation answers, and a provider model update can jump the numbers in a day for reasons that have nothing to do with you — which is exactly why the log in How to Track Your Brand Mentions in ChatGPT, Gemini and Perplexity pins dates and versions. When the internal trend is stable, add the competitive view with Measuring Share of Voice in AI Answers.
Keep reading
- What Is Answer Engine Optimization (AEO)? — the definitions and the founder's three steps.
- AEO vs SEO: What Changes and What Stays — why entity consistency carries more weight here.
- The AI Answer Audit Checklist — find the contradictions before you publish anything new.