AI Quick Summary: ChatGPT knows a brand through two paths: live search (crawling the web while writing) and model knowledge (your footprint in training data). Visibility needs both: extractable passages for the search path, consistent entity definitions for the memory path — measured with the same frozen question set in 30-60-90 day rounds.
You ask ChatGPT about your brand and you're absent from the answer. Your competitor is there. This article explains why that picture forms and what needs doing in the first 30 days.
First the Mechanism: Two Separate Paths
ChatGPT knows a brand through two paths. The first is live search: the model crawls the web while writing its answer and sources the passages it finds. The second is model knowledge: your brand being defined in training data and context consistently with its category.
The two paths answer to different signals. Your site can rank first on Google yet never appear on ChatGPT — because ranking and answer production are different jobs. The practical rules of both paths are in our two-path article.
Why Is Your Competitor There While You're Not?
Start the diagnosis with three questions; the answers reveal the missing signal:
On which questions does your competitor appear? Brand-name questions, category questions, or comparisons? Brand-only means an awareness gap; category questions mean a content gap; comparisons mean a proof gap.
On which path does your competitor win? Source links in answers mean the search path (a passage gap); fluent linkless mentions mean the memory path (an entity gap). The path-detection method is in our two-path article.
What does your competitor have that you lack? Definition-first page intros, consistent directory profiles, press repetition, structured data? Reduce the gap to concrete items; "they're a big brand" is not a diagnosis. In most categories the gap is 3–5 concrete items and closes in 90 days.
These three answers set which of the 5 steps below gets the weight. Do the diagnosis in writing: table each competitor's questions and paths. The table turns "competitors are better than us" from a feeling into "competitors lead on these 4 items" — and information turns into a task list. Update this table every round; closing rows are your motivation, widening rows your alarm.
Step 1: Never Start Without Measuring
"We're invisible" is a feeling, not a diagnosis. Diagnosis needs your position on your category's real buyer questions: on how many are you mentioned, on how many sourced, where are competitors?
A Citation Baseline answers exactly this: a current-state report within 5 business days. Every program started without measurement spends resources without knowing which gap it's working on.
Building your own panel follows a known skeleton: 20 questions (category definition, comparison, purchase intent, problem-solving), 4 models, a frozen set, 30-60-90 rounds. Set construction details are in the unbranded question-set guide; citation rate and Citation Delta definitions in the glossary.
Step 2: Make Your Passages Extractable
ChatGPT pulls the question-matching section, not the whole page. The unit under evaluation is the page no longer — it is the passage. The rule is simple: does a section give a complete answer even when read without the rest of the page?
A paragraph that fails:
Among the approaches described above, this method stands out and generally produces good results.
Which approach, stands out where, good by what measure? Meaningful inside the page, worthless when extracted. The fix is usually re-sectioning, not rewriting: see the quotable passage rules.
Repair order: first convert your 10 most valuable pages' intros to definition-first form, then add standalone 40–80 word summary passages to each page, then date the undated pages. These three moves produce first-round movement on most search-path questions.
Step 3: Define Your Entity
The model treats brands not as websites but as concepts with attributes and relationships — entities. A clear map of brand, product, people, location, and proof strengthens this recognition. Inconsistent or missing entity information means the model skips you for a more clearly defined competitor.
Entity definition has three layers:
- Identity layer. Name, category, location, founding facts — verbatim identical on every platform.
- Offer layer. What you sell, to whom, in what range — concrete and current.
- Proof layer. References, numbers, dates — the facts carrying your claims.
Gather all three layers in one document (the master copy) and compare against 10 critical platforms every quarter. When you find a gap, fix the platform; the master copy is the single source of truth.
Step 4: Add Third-Party Proof
What you write on your own site is a claim; what independent sources write is proof. Sector publications, reviews, and references enter the model's trust math. One strong sector source outweighs ten weak blog posts.
Proof priority order: first consistent information on sector directories and profile pages (cheap, fast), then natural repetition in customer reviews (medium cost, cumulative), then press and third-party reviews (expensive, but they produce the proof that enters reasoning). Each layer appears as a separate step on the Citation Delta curve.
The measure in proof work is repetition: the same core facts must pass in different sources' own sentences. Not copy-paste press releases — independent repetitions produce trust.
Step 5: Re-Measure in 30 Days
Visibility work is never a one-off. 30 days after the first measurement, re-run the same question set: on which questions did you advance, on which are you stuck? Stuck questions are the next round's task list.
Round discipline: same questions, same conditions, same counting rules. Never change questions mid-round, add or drop models, or update the counting rule between rounds. The full discipline is in our measurement methodology. The 90-day evidence is in the case study: Filmfolk went from 0% to 81% in this order.
The First 30-Day Calendar
Spread the five steps across weeks:
Week 1: Diagnosis. Answer the competitor gap's three questions; build the 20-question panel; take your Baseline measurement.
Week 2: Passage repair. Convert your 10 most valuable pages' intros to definition-first form; add summary passages; refresh dates.
Week 3: Entity and schema. Write the master-copy document; fix platform gaps; add missing structured data.
Week 4: Proof and round prep. Update directory profiles; plan the first press/review moves; put the day-30 round on the calendar.
This calendar runs on one person's half-time work; with a team, parallelize — but manage round discipline single-handedly.
Anatomy of a Searched Answer
Take ChatGPT's sourced answer apart; each piece tests different content of yours:
- Answer body. The summary the model writes in its own sentences. Entering here needs definition-first passages.
- Source list. The sites the answer rests on. Listing here needs quotable, dated pages.
- Follow-ups. "Ask this too" links. Entering here needs topic authority — one page won't do, a cluster will.
Log these three pieces separately in your panel: did you pass in the body, are you on the list, are you in the suggestions? They are three different maturity levels; most brands enter the list first, then the body, last the suggestions. Read your progress in that order.
Team and Budget: Who, How Long?
The first 30 days' workload is known: panel setup (1 day), 10-page passage repair (3–5 days), master copy and platform fixes (2–3 days), schema additions (1–2 days), proof planning (1 day). Two weeks' work for one person; a week with a content team.
Budget items: in-house team hours (the biggest item), Baseline measurement (free), directory and profile updates (usually free), press/review work (variable). Working with a consultant? Clarify scope with the proposal comparison frame. The measure is simple: if in-house cost passes the consultant fee in 3 months, buy it externally.
3 Common Mistakes
- Settling for SEO metrics. Rankings can rise while answer visibility sits at zero; the two get measured separately.
- Producing generic content. "Everything about AI" articles pull nobody; question-specific, standalone passages are needed.
- Watching one model. A brand visible on ChatGPT can be invisible on Perplexity. Track all four models together.
A fourth mistake exists: changing without measuring. Update content without re-running the same questions and you'll never know what worked. The same panel runs before and after every change.
Next Step
With five steps done, the next job is building the routine: monthly page maintenance, quarterly entity audits, 30-60-90 rounds. Start with a Citation Baseline to learn where you stand; the full method is in the Adaptive RAG methodology. General questions are answered on the FAQ page. Read your first round's results together with this article's frame on day 30 — advancing questions prove your discipline, stuck questions prove next round's task list. Once the routine is built the work lightens: monthly maintenance, quarterly audits, regular rounds. Six months from now, looking back, you'll see your answer-layer place was no accident but this article's five steps compounded. Start with measurement, continue with passages, sustain with rounds. Discipline quietly beats scattered talent every round.
FAQ
Do I need to advertise to appear on ChatGPT?
No. ChatGPT answers can't be bought with ads; the model picks quotable content and consistent entity information. Visibility is content and measurement work, not media budget.
Where does ChatGPT get brand information?
From two sources: current web pages on the live-search path, and your footprint in training data on the model-knowledge path. Both sources need feeding.
How many days until ChatGPT visibility changes?
Weeks on the search path, months on the memory path. The same questions re-run 30 days after the first measurement; the meaningful curve reads on days 60–90.
Related Articles
ChatGPT Live Search vs Model Knowledge
What changes in practice between the two paths.
How to Write Citable Passages
Sections that stand when ripped from context.
How to Build an Unbranded Question Set
The 20-question skeleton behind every measurement.
What Is a Citation?
Citation, mention, and ranking are three different things.
About the author
Bora Kurum is the founder of RAG Signal and a Ph.D. researcher at Istanbul Bilgi University, where his work focuses on LLM retrieval behavior and brand representation in generative AI systems. Read more →

