Does changing your domain name wipe out your AI recommendations?

You moved your product to a new domain last month. When you type your category into ChatGPT, your old brand name appears, or worse, your competitors show up while your new domain is missing entirely. It feels like your brand authority vanished during the move.

The reality is that a domain migration does not erase your AI recommendations instantly, but it does split your brand identity across two separate systems inside an AI engine. One system lives in static model weights. The other system lives in real-time search retrieval.

Understanding how these two systems handle your brand migration determines what you should fix and what you must simply wait out.

Parametric memory versus live answer crawlers

Large language models process brand entities through two distinct mechanisms: parametric memory and retrieval-augmented generation.

Parametric memory consists of the static weights built during model training. Training crawlers visit web pages to construct this dataset months before a model goes live. GPTBot visited your old domain long before your rebrand. ClaudeBot and Google-Extended compiled text from your former website into training batches that are now frozen inside model weights. In parametric memory, your old domain is still the authority for your brand because the model learned that fact in the past.

Retrieval-augmented generation works differently. When a user asks an engine for software recommendations today, the engine uses an answer crawler to fetch current web pages. Answer crawlers behave like live web browsers working on behalf of an active prompt session.

ChatGPT-User visits a web page when a prompt triggers a live web search in ChatGPT. Claude-SearchBot performs the same task for Claude, while Perplexity-User and OAI-SearchBot handle live fetches for their respective engines.

If your old domain uses proper 301 redirects, an answer crawler like ChatGPT-User that follows a link to your old domain will land on your new domain. The retrieval system reads the current content, sees that the product is identical, and reconciles the entity during that specific session.

The point of failure occurs when answer crawlers are blocked or redirected incorrectly. In our index of 54,082 domains updated on August 8, 2026, 33,670 domains returned a readable robots.txt file. Out of those readable domains, 5,497 block at least one AI agent, and 2,576 block at least one answer crawler.

Among those 33,670 readable domains:

• GPTBot (a training crawler) is blocked by 5,080 domains.
• ClaudeBot (a training crawler) is blocked by 4,603 domains.
• Google-Extended (a training crawler) is blocked by 4,275 domains.
• PerplexityBot (a search crawler) is blocked by 2,147 domains.
• ChatGPT-User (an answer crawler) is blocked by 2,074 domains.
• OAI-SearchBot (an answer crawler) is blocked by 1,579 domains.
• Perplexity-User (an answer crawler) is blocked by 1,402 domains.
• Claude-SearchBot (an answer crawler) is blocked by 1,400 domains.
• Claude-User (an answer crawler) is blocked by 1,384 domains.

Blocking a training crawler like GPTBot prevents OpenAI from using your new domain in future foundation models. Blocking an answer crawler like ChatGPT-User prevents ChatGPT from verifying your new domain during live web searches today.

Even among technical teams, crawler blocking is common. In our survey of 4,226 active Y Combinator companies conducted on August 7, 2026, 3,755 domains had readable robots.txt files, and 253 blocked at least one AI crawler. If your new domain blocks answer crawlers, live retrieval cannot bridge the gap between your old parametric presence and your new web address.

A side by side test for your migration

To diagnose where your brand stands after a domain migration, run side by side tests in an engine that supports optional web browsing.

Open a fresh chat session and disable web search. Run two distinct prompts:

1. "Tell me about [Old Brand Name]."
2. "Tell me about [New Brand Name]."

Without web search, the engine relies entirely on parametric memory. In most cases, the model will output detailed information for your old brand name and display confusion or hallucination for your new brand name. This confirms that the existing model weights were trained before your migration took place.

Next, open a new chat session and enable web search. Run the same two prompts, along with a category discovery prompt:

1. "Tell me about [Old Brand Name]."
2. "Tell me about [New Brand Name]."
3. "What are the top software options for [Your Specific Niche]?"

Watch the search queries generated by the engine. If the answer crawler works correctly, a search for your old brand name will hit your old domain, follow the HTTP 301 redirect to your new domain, and summarize your new site accurately.

If search is enabled but the engine still returns outdated information or omits your brand, inspect your server logs for the answer crawler user agents listed above. A misconfigured firewall, a CDN rule, or a restrictive robots.txt entry on your new domain may be dropping answer crawlers before they complete the retrieval step.

Why model weights take months to catch up

A successful domain migration transfers your live search visibility, but it cannot force an immediate update to frozen model weights.

Foundation models are re-trained periodically on massive infrastructure runs. The training pipeline from data collection to final weight freezing takes months. No tool, vendor, or API call can speed up this cycle for your brand.

During this transition period, AI engines exhibit split behavior. When a user asks an ungrounded, general question without live search, the model draws on parametric memory and may cite your old domain or omit your brand. When a user asks a query that triggers live retrieval, the engine reads your redirected pages and cites your new domain.

Entity resolution inside LLMs gradually consolidates these identities as three things happen over time:

1. Third-party sites update their references and backlinks from your old domain to your new domain.
2. Web search indexers crawl and record the standard 301 redirect paths across your site.
3. Model providers execute new training runs using datasets collected after your migration.

Until a new model version is trained and deployed by the model provider, your parametric presence will remain anchored to your historical brand identity.

What not to buy during a domain move

Because domain migrations create visible gaps in AI answers, agency services have emerged promising to fix or fast-track AI brand transfers. Most of these offerings exploit misunderstanding of how LLMs process data.

Do not buy services that promise to re-index your brand inside ChatGPT, Claude, or Gemini. AI lab vendors do not offer on-demand model re-training endpoints for private web domains.

Do not buy domain redirect tools marketed specifically for AI engines. A standard HTTP 301 redirect configured at your DNS or web server level handles answer crawlers like ChatGPT-User and Claude-SearchBot identically to standard web browsers.

Do not buy schema markup packages or llms.txt files under the assumption that they accelerate entity consolidation or improve AI citations. Schema markup provides JSON-LD metadata intended for search engine rich snippets, but answer crawlers parse plain web page text and do not prioritize schema entities over body content. An llms.txt file is simply a curated text file listing links for developers; answer crawlers do not read it to verify domain ownership or update entity mappings.

Focus instead on standard web operations: maintain permanent 301 redirects from every old URL path to its new equivalent, verify that answer crawlers are explicitly allowed in your robots.txt file, and update high-authority web mentions where you control the listing.

Tracking your post-migration baseline

Measuring AI recommendations requires repeated testing because non-deterministic models do not yield identical answers on every run. Single prompt checks often give a false impression of progress or loss.

At Standing, we track recommendations across five engines: ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews. We execute a fixed prompt basket five times per question per engine and calculate the resulting recommendation rate using a Wilson score interval. A single run showing 34 citations out of 100 queries is expressed with its margin, such as 34, plus or minus 6 out of 100 runs within a 95% confidence interval.

Competitors typically run each prompt once, producing single numbers that hide natural engine variance. Single runs make it impossible to tell whether a missing mention is caused by a broken migration path or standard output sampling.

We offer three plans to track entity visibility over time: Track at $100 per month for 3 domains, Optimize at $300 per month for 5 domains with custom prompt configurations, and Agency at $500 per month for 50 domains with automated monthly re-scans.

If your redirects are clear and your answer crawlers are unblocked, live recommendations will continue through live retrieval while parametric memory steadily catches up in future model cycles.