How long it takes to change what AI says about you

Someone on your team ran a prompt in ChatGPT, noticed your company was missing from the recommendations, and forwarded the screenshot to leadership. Shortly after, an agency promised they could get your brand cited across AI engines within 30 days.

That 30-day timeline is a convenient fiction. It takes a single real-time search clock and presents it as the timeline for the whole system. In reality, AI engines rely on three separate systems operating on radically different schedules. If you do not understand which clock you are trying to move, you will spend time and money targeting mechanisms that cannot respond on the timeline you expect.

The three distinct clocks governing AI answers

AI engines do not possess a single memory bank. When an engine generates a response, it combines static knowledge baked into its weights with live content fetched from the web. Influencing these outputs requires operating against three distinct clocks.

Clock 1: Live search retrieval (3 days to 2 weeks)

When an engine decides a prompt requires current information, it delegates the query to a search component. An answer crawler fetches live web pages, extracts context, and feeds that content directly into the prompt context window.

If a source publication updates a page today, and an answer crawler fetches it tomorrow, the change can appear in generative answers within days. This is the fastest clock. It applies only when an engine chooses to browse the web for a specific query.

Clock 2: Source publication cycles (2 weeks to 3 months)

Answer engines rely heavily on third-party sources: industry roundups, review aggregators, trade publications, and comparison blogs. An AI model rarely cites your website directly for broad commercial queries. It cites the authoritative sources that evaluate your category.

To change what answer crawlers find, you must first change what those underlying sources say. Securing a inclusion on an industry comparison list or getting a trade outlet to update a category review depends on external editorial schedules. This clock moves at the pace of traditional media and web content management.

Clock 3: Model weight updates (6 months to 18 months)

When an engine answers a prompt without browsing the web, it relies entirely on its parametric memory. These are the static weights generated during massive pre-training and fine-tuning runs.

Changing a model's parametric memory requires your brand to exist in the massive text corpora gathered by training crawlers before a training run begins. After the dataset is collected, the model must be trained, evaluated, safety-aligned, and deployed. A brand visible in live web results today may remain entirely absent from parametric responses for a year or longer.

Why 30-day agency promises are misleading

Vendors who claim they can optimize your AI presence in 30 days are manipulating the fastest clock.

They do this by identifying prompts that consistently trigger real-time search retrieval. They publish optimized content on quickly indexed pages or modify live site structures, then run specific queries that force the engine to browse the web. When the engine fetches the new content and mentions the brand, the vendor declares victory.

This creates an illusion of permanent visibility. As soon as the user alters the query slightly, or as soon as the engine decides not to invoke web search for that prompt, the answer reverts to parametric memory. Your brand disappears again because Clocks 2 and 3 were never addressed.

Furthermore, technical optimizations on your own domain have strictly limited reach. Some marketers assume adding structured data or deploying an llms.txt file will speed up inclusion. We have seen no evidence that schema markup or llms.txt files improve citation frequency in AI engines, and relying on them as shortcut tactics wastes valuable effort.

Answer crawlers versus training crawlers

Understanding the difference between answer crawlers and training crawlers is critical when auditing your visibility or troubleshooting why a domain is missing.

Training crawlers gather bulk text to build future foundation models. Blocking or unblocking a training crawler changes whether your site enters the dataset for a model release next year. Answer crawlers fetch live pages to construct immediate responses to user queries. Blocking an answer crawler stops an engine from citing your site today.

Many organizations misconfigure their web servers because they fail to distinguish between these two crawler types. In our crawler index snapshot from August 8, 2026, we analyzed a panel of 54,082 domains. Of those, 33,670 returned a readable robots.txt file.

Across those 33,670 readable domains, 5,497 block at least one AI agent. However, only 2,576 block at least one answer crawler. Thousands of sites are actively blocking training crawlers while leaving answer crawlers unrestricted, or vice versa, often without understanding the impact on their visibility timeline.

The breakdown across specific agents among the 33,670 readable domains illustrates how widespread these configurations are:

• GPTBot (training): blocked by 5,080 of 33,670 domains
• ClaudeBot (training): blocked by 4,603 of 33,670 domains
• Google-Extended (training): blocked by 4,275 of 33,670 domains
• PerplexityBot (search): blocked by 2,147 of 33,670 domains
• ChatGPT-User (search): blocked by 2,074 of 33,670 domains
• OAI-SearchBot (search): blocked by 1,579 of 33,670 domains
• Perplexity-User (search): blocked by 1,402 of 33,670 domains
• Claude-SearchBot (search): blocked by 1,400 of 33,670 domains
• Claude-User (search): blocked by 1,384 of 33,670 domains
• Googlebot (other): blocked by 610 of 33,670 domains

This split is evident even among venture-backed technology companies. In our census of 4,226 active Y Combinator companies conducted on August 7, 2026, 3,755 returned a readable robots.txt file. Of those readable domains, 253 block at least one AI crawler.

If you unblock an answer crawler like ChatGPT-User or Perplexity-User, you allow live web search to reference your pages within days. If you unblock a training crawler like GPTBot or ClaudeBot, you will see zero immediate impact on user queries today.

Measuring changes in non-deterministic systems

Because AI models are non-deterministic, measuring whether your visibility has actually changed is technically difficult.

If you run a prompt once in ChatGPT and see your brand, that single output is not a measurement. The engine might return your brand on one run and omit it on the next four runs due to temperature sampling and dynamic context assembly. Stating that your brand has a specific ranking or fixed score based on one attempt is meaningless.

To establish a valid baseline, you must sample prompts repeatedly over time. At Standing, our method uses a fixed basket of industry prompts, running each prompt five times per question across five major engines: ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews.

We apply a Wilson interval to the results to account for model variance. A brand score should never be expressed as a single static integer. It must be stated with a confidence band, such as a visibility score of 34, plus or minus 6 out of 100, calculated within a 95% confidence interval.

Most optimization vendors run a prompt a single time, record a positive match, and report it as a permanent gain. Competitors avoid publishing confidence bands because single-run testing hides the underlying instability of AI citations.

Tracking changes across realistic timeframes requires structured monitoring:

• Track Tier ($100/month): Covers 3 domains across all 5 engines using standard industry prompt baskets.
• Optimize Tier ($300/month): Covers 5 domains with custom prompt configurations to track specific product categories.
• Agency Tier ($500/month): Covers 50 domains with automated monthly re-scans to measure long-term trajectory across broad portfolios.

When you adjust your strategy, expect real-time retrieval metrics to shift over weeks as third-party sources pick up your brand. Expect parametric visibility to remain flat for months until model providers complete their next major training and alignment passes. Anyone offering a guaranteed timeline or promising a 30-day turnaround across all AI answers is selling a fundamental misunderstanding of how these models work.