What to fix first when AI has never heard of you

You ran a scan across ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews, and your brand score came back at 0, plus or minus 0. You spent thirty minutes asking ChatGPT for recommendations in your product category, and it named three competitors while ignoring your company entirely.

Do not buy a five thousand dollar monthly retainer from an agency promising guaranteed AI search rankings. Do not waste an afternoon adding schema markup or uploading an llms.txt file to your web server. Schema markup and llms.txt files do not increase your citation rates in AI engines, and any vendor promising guaranteed placement in a non-deterministic system is selling a myth.

When an AI engine has never heard of your brand, fixing the problem follows a strict sequence ordered by cost and speed to impact. You start by opening access to answer crawlers, move to writing liftable sentences on your own domain, and finish by earning mentions on the third-party web pages that the engines already trust.

Unblock answer crawlers before doing anything else

Checking crawler access costs nothing and takes five minutes in your web server configuration. If an AI engine cannot fetch your pages when a user asks a question, your content does not exist for that session.

Many marketing teams conflate training crawlers with answer crawlers, leading to costly mistakes. A training crawler scrapes web pages to gather dataset bulk for training future foundational model weights. Blocking a training crawler affects whether your site enters a future model update months down the line. An answer crawler performs live web retrieval while a user is actively asking a prompt. Blocking an answer crawler stops an engine from reading your site right now during a active query.

In our crawler index update from 2026-08-08, we checked a panel of 54,082 domains. Of those, 33,670 returned a readable robots.txt file. A total of 5,497 of 33,670 readable domains blocked at least one AI agent, and 2,576 of 33,670 readable domains blocked at least one answer crawler.

This issue occurs frequently even at technical companies. In our census of 4,226 active Y Combinator companies conducted on 2026-08-07, 3,755 returned readable robots.txt files, and 253 of those 3,755 readable domains blocked at least one AI crawler.

Inspect your robots.txt file immediately for blocks targeting answer agents:

• ChatGPT-User (blocked by 2,074 of 33,670 readable domains)
• PerplexityBot (blocked by 2,147 of 33,670 readable domains)
• OAI-SearchBot (blocked by 1,579 of 33,670 readable domains)
• Perplexity-User (blocked by 1,402 of 33,670 readable domains)
• Claude-SearchBot (blocked by 1,400 of 33,670 readable domains)
• Claude-User (blocked by 1,384 of 33,670 readable domains)

If your robots.txt file blocks these agents, search-enabled runs in ChatGPT, Perplexity, or Claude cannot pull data from your domain when a user asks for product recommendations.

Do not confuse these answer crawlers with training crawlers like GPTBot (blocked by 5,080 of 33,670 readable domains), ClaudeBot (blocked by 4,603 of 33,670 readable domains), or Google-Extended (blocked by 4,275 of 33,670 readable domains). Unblocking GPTBot allows OpenAI to scrape your site for future base models, but it will not fix your visibility in ChatGPT today. Unblocking ChatGPT-User and OAI-SearchBot addresses real-time retrieval immediately.

For context, Googlebot is blocked by 610 of 33,670 readable domains in our index. Far more domain owners block AI search agents than traditional search crawlers, often due to default web application firewall rules or copied configuration templates.

Write liftable sentences on your own domain

Once answer crawlers can fetch your pages, you must give them content that retrieval systems can easily extract and reassemble into an answer.

When an engine executes a live search to answer a user prompt, it searches for explicit, factual declarations. If your homepage relies on abstract marketing language, the retrieval model cannot find plain claims to support a recommendation.

You need pages on your website that answer concrete buying questions in liftable prose. A liftable sentence is a straightforward factual claim that uses clean subject-verb-object structures. It states unambiguously what your software does, who uses it, what technical environments it supports, and how much it costs.

Avoid indirect phrasing. Do not write that your platform empowers modern teams to accelerate enterprise workflows through seamless integrations. Write that your software is a web-based project management tool for architectural firms, starting at forty dollars per user per month. The second sentence gives a retrieval model exact facts to extract when a user asks for project management software built for architects.

Do not rely on structural metadata to fix poor prose. Structured schema markup and llms.txt files do not improve your citation rates in major AI engines. Our own evaluations show no evidence that adding schema or llms.txt files increases brand mentions. Search engines read the plain text rendered on your pages. If your visible prose lacks clear buying facts, metadata files will not bridge the gap.

Earn mentions on the sources engines already cite

Publishing liftable content on your site is necessary, but owned content alone is rarely enough. The third stage of this process is the slowest and most valuable part of the work: securing presence on external domains that AI search engines already reference for your category.

When ChatGPT or Perplexity answers a buying query, it does not rely solely on vendor websites. It queries external reference sources to evaluate category consensus. It searches for review aggregators, editorial roundups, comparison blogs, industry news outlets, and active forum discussions.

If an engine consistently retrieves search results from five specific comparison blogs or directory listings when answering questions in your niche, those five pages form the baseline knowledge for the query. If your brand is absent from those external pages, the engine has no third-party validation to justify citing your business.

Identify the exact web pages that engines cite when responding to prompts in your space. Look at the reference links provided at the bottom of responses in Perplexity, ChatGPT, and Google AI Overviews. Reach out to those external publishers, pitch inclusion in their product roundups, update your profiles on industry software directories, and participate in technical forums where users ask for software suggestions.

This work takes time. Editorial teams take weeks to update articles, search engines re-index external sites on their own schedule, and model retrieval systems update gradually over time. Nothing in AI engine optimization happens in days. Vendors promising instant inclusion across AI engines do not understand how non-deterministic retrieval systems operate.

Track visibility using proper statistical intervals

Because generative AI engines are non-deterministic, checking a prompt manually once in ChatGPT gives a misleading picture. An engine might fail to cite your brand on one attempt, then cite it on four subsequent runs with the exact same prompt, or vice versa.

Competitor tracking tools often run each query a single time to save computing costs. This produces single-number metrics that swing wildly from day to day without reflecting actual shifts in visibility.

Standing tracks visibility across five engines: ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews. Our method uses a fixed basket of prompts run five times per question per engine. We process those results with a Wilson interval to publish a mathematically grounded score with clear upper and lower bounds, such as 22, plus or minus 4, within a 95% confidence interval.

Running every prompt five times requires significantly more compute, which is why competitors avoid doing it and fail to provide confidence bands. However, without a confidence interval, you cannot distinguish between normal model variance and actual progress resulting from your work.

If your initial baseline scan shows a score of 0, plus or minus 0, tracking these confidence bounds lets you measure genuine, statistically valid improvements as your unblocking, content updates, and third-party placement work begin to take effect. Standing offers Track at $100 per month for 3 domains, Optimize at $300 per month for 5 domains with custom prompt baskets, and Agency at $500 per month for 50 domains with a monthly re-scan.