ChatGPT is where a growing share of buying research now begins. We fix the crawler access, entity signals, and source credibility that determine whether it names your brand or a competitor's.
ChatGPT sometimes searches the live web and cites what it finds. Other times it answers from what the model already carries — no search, no citations, just a recommendation stated from memory. Which path it takes depends on the question, the model version, and settings you don't control.
That means optimizing for one path isn't enough. Live retrieval rewards crawler access, rendering, and structure. Model memory rewards consistent, corroborated presence across the web over a long period. The brands that do well in ChatGPT tend to have both.
The good news is the first path is fixable quickly and it's where most brands are failing outright. A large share of the sites we audit are simply blocking GPTBot, often through a robots.txt rule or bot-management setting nobody remembers adding.
For ChatGPT specifically, this sequencing question has a fairly clear answer for most brands.
Best for: Almost every brand, first. If GPTBot is blocked, nothing else you do for ChatGPT matters much.
Best for: Established brands defending category position once retrieval is already clean.
Four factors, roughly in the order they tend to break.
Whether GPTBot and OAI-SearchBot receive a 200 with full content. This is the most common failure and the easiest to fix — often a single robots.txt line.
Whether the HTML response contains your actual content or an empty shell awaiting JavaScript. Access without rendering is still invisibility.
Whether ChatGPT connects your pages to a recognised company with defined attributes, or treats your brand name as an ambiguous string.
Whether independent sources describe you consistently. ChatGPT's source selection leans heavily on corroborated information over single-source claims.
Access remediation, entity work, content structuring, and the ongoing testing that catches when ChatGPT starts describing you differently.
Auditing and fixing robots.txt, WAF rules, rate limits, and bot-management settings so OpenAI's crawlers receive full content rather than a block or a challenge.
Confirming the HTML response contains your content, and specifying server-side rendering or pre-rendering where it doesn't.
Schema, naming consistency, and sameAs corroboration so ChatGPT resolves your brand to a specific company rather than an ambiguous name.
Restructuring priority pages so a clean, self-contained passage can be lifted and attributed without distortion.
Earning consistent independent references, which disproportionately influence what ChatGPT states confidently versus hedges on.
Monthly testing across a fixed prompt set, tracking whether you're named, how you're described, and which competitors are gaining.
How ChatGPT handles a question varies a lot by category, which changes where the effort goes.
ChatGPT usually searches for software comparisons and current pricing, since it knows this information dates quickly. Retrieval work matters most here.
Product recommendations mix memory and retrieval. Structured product data matters for the retrieval path; brand familiarity drives the memory path.
'Who should I hire for X' questions often answer from memory rather than searching, which makes long-term corroboration disproportionately important.
ChatGPT leans heavily on documentation for technical questions. Docs that aren't crawlable are a common and costly gap.
Retrieval first because it's fast and usually broken, then entity and corroboration work underneath.
We verify what GPTBot and OAI-SearchBot actually receive, and benchmark how often ChatGPT currently names you across a real prompt set.
We test systematically how ChatGPT describes your brand and products, and trace every error back to its probable source.
Access issues resolved first, then schema and structural work, with corroboration building running underneath on a longer timeline.
Fixed prompt set re-run monthly, tracking citation share, description accuracy, and competitor movement as models update.
Almost everything that improves ChatGPT visibility — crawler access, rendering, entity clarity, corroboration — improves Perplexity, Gemini, Claude, and Copilot simultaneously. Platform-specific tactics are a smaller share of the work than most agencies imply.
The questions clients ask us most before starting. If yours isn't here, ask us directly on a consultation call.
Check your robots.txt for GPTBot and OAI-SearchBot directives, and check whether your WAF or bot-management tool blocks unfamiliar agents by default. Then fetch a page as GPTBot and compare the response against what a browser renders.
Blocking is common and often unintentional — frequently added years ago as a blanket AI-crawler rule, or by a security tool nobody associated with marketing.
It's a business decision. Allowing it is a precondition for being cited when ChatGPT searches. Blocking it protects your content from being used without attribution.
Most brands selling a product or service benefit from being findable. Publishers monetising content directly have a legitimate case for restricting. We'll walk through the tradeoff rather than assuming.
GPTBot is documented as being used for training. OAI-SearchBot is documented as serving search and citation. They can be controlled separately in robots.txt, so you can allow search retrieval while disallowing training crawls.
That distinction matters to a lot of clients and it's worth configuring deliberately rather than treating all AI crawlers as one decision.
Usually one of three reasons: it can't access your site, it can access but can't resolve your brand as a credible entity, or your competitor has substantially more independent corroboration.
Benchmarking tells you which. In our experience the first is far more common than brands expect, and it's the cheapest to fix.
Not directly, and anyone claiming otherwise is misrepresenting it. What you can do is influence the inputs: make accurate information accessible, keep descriptions consistent across every property, and build corroboration for the facts you want stated.
Errors that come from model memory rather than retrieval are the slowest to correct, since they depend on the wider web being updated.
Access fixes can show within two to eight weeks depending on crawl cycles. Entity and structural work typically four to eight weeks. Memory-driven changes take quarters.
We report monthly so you can see which pathway is responding rather than waiting on a single distant outcome.
Roughly 80% is shared — access, rendering, entity clarity, corroboration. The differences are that Perplexity searches far more consistently, cites more visibly, and moves faster, while ChatGPT mixes retrieval with memory more heavily.
In practice we optimize for both simultaneously rather than running separate programmes.
A fixed prompt set of 40 to 60 category questions, run multiple times per cycle to capture variance, recording whether you're named, how prominently, how you're described, and which competitors appear.
We report the trend and the variance rather than a screenshot, because single outputs are genuinely unreliable evidence.
We'll test whether OpenAI's crawlers can read your site, benchmark how often ChatGPT names you across your category's real questions, and show you who it recommends instead.
No commitment · 45 minutes · Immediate value
We’ll run your category’s real questions live on the call and tell you honestly whether this is worth your budget. No pitch deck.