Academy · Updated 2026-08-23 · 7 min read
How to get your brand recommended by ChatGPT
Customers keep finding competitors through AI recommendations. The playbook, in the order that works: measure, read why you lose, enter the evidence pool, close the content gaps, re-measure.
When a buyer asks ChatGPT "what's the best X" and it names three brands, that answer was assembled in seconds from two inputs: what the model already associates with the category, and — when it searches — whatever pages it pulled into context. Getting recommended means winning one or both of those inputs. Neither responds to wishing, and both respond to specific, checkable work.
Step 1 — Measure what the engines say today
Write down the questions your buyers actually ask — problem-led, best-of, comparison, category — and put them to the engines with live search on, several times each, storing every answer. The split that matters: questions that name your brand versus questions that don't. Branded questions test recall; unprompted questions test whether you exist to a cold buyer. Expect the unprompted number to be brutal; it usually is.
We publish our own numbers because we ask you to trust ours. On 2026-08-23 we ran the full AEOSearch audit on AEOSearch itself — 300 stored, grounded answers across ChatGPT, Claude, Gemini, Grok and Perplexity. Our unprompted mention rate was 0%. Our citation share was 0%. That is the honest starting line for a young brand, and every method on this page is what we are running on ourselves, in public, to move it.
Step 2 — Read why the competitor wins, question by question
Every lost question has a mechanism, and the stored answer shows it: the engine cited a page you don't have (content gap), leaned on a roundup that doesn't include you (authority gap), pulled a fresher page (freshness gap), or lifted a competitor's structured, extractable definition (format gap). The winning source is right there in the citations. Diagnosis is reading, not guessing — this is the audit's gap diagnosis step.
Step 3 — Get into the evidence pool
A grounded engine can only cite pages it can crawl. Before any content work, verify the prerequisites: AI crawlers not blocked in robots.txt, schema.org structured data present, an llms.txt published, and — the hard one — your domain actually appearing among cited sources in your category's answers. The AI crawler & citation checker walks through each check with the exact user-agent strings.
The authority half is slower: the engines lean heavily on third-party roundups and comparison pages when recommending tools and products. Being absent from the roundups your category's answers cite is a structural handicap no owned page fully offsets — which is why the fix plan treats "pitch the roundups that engines actually cite" as work, with the specific URLs your audit surfaced.
Step 4 — Close the content gaps, in the shape engines lift
Write the canonical page for each question you lose: definition in the first forty words, FAQ blocks with self-contained answers, comparison tables an engine can lift whole, structured data marking it up, and a visible updated date. Not because these are magic — because the stored answers show these are the page shapes engines already cite for your questions.
Step 5 — Re-measure, and watch one number
Re-run the same question set monthly, same sampling, and diff against the stored baseline. For a brand starting from zero the headline metric is time to first unprompted mention; after that, citation share — how often your own domain is among the sources. Movement you can't show in a diff of stored answers didn't happen. Start with the free scorecard; it becomes the baseline everything else is measured against.
Questions, answered straight
How does ChatGPT decide which brands to recommend?
From two inputs: the associations in its training data, and — when it searches — the pages it pulls into context at answer time, which lean heavily on roundups, comparison pages, and canonical explainers. Getting recommended means appearing in those cited sources and in the category's written record, not optimizing a homepage.
How long does it take to get mentioned by AI assistants?
Honest answer: it varies, and anyone quoting a fixed timeline is guessing. Evidence-pool fixes (crawler access, schema, citable pages) can affect grounded answers within weeks because those answers are built from live search; training-data associations move on model-release timescales. Measure monthly and track time to first unprompted mention rather than trusting a promise.
Why do competitors show up in AI answers and my brand doesn't?
The stored answers usually show one of four mechanisms: the engine cited a page type you don't have, leaned on a roundup that omits you, pulled a fresher source, or lifted a competitor's structured, extractable content. Reading the citations in the answers you lose turns this from speculation into a specific fix list.
Can I pay to be recommended by ChatGPT?
No. There is no placement to buy in organic assistant answers, and any vendor promising guaranteed mentions is promising something the engines don't sell. What is buyable is measurement and execution: knowing exactly which questions you lose, why, and shipping the fixes — then proving movement against a stored baseline.
— The AEOSearch team