More than a year after Contentsquare officially completed its merger with Hotjar in July 2025, ChatGPT still recommends Hotjar as an independent heatmapping product — with no mention of Contentsquare. That finding comes from Semrush, which tested how major brands appear in ChatGPT answers and documented a pattern it calls the "LLM positioning lag": a company changes its positioning, but large language models keep repeating the old one.

The stakes are quantified. According to Capgemini, the share of consumers who replaced search engines with generative AI for product and service recommendations grew from 25% in 2023 to 58% in 2025. A growing majority of prospects now form their first impression of a brand from an AI answer — and if that answer describes who the company was, not who it is, the prospect may never visit the website to correct the record. With agentic commerce emerging, where AI agents make purchases on users' behalf, inaccurate brand data can also translate directly into lost sales.

Three causes, all tool-signal analysis

Semrush identifies three mechanisms behind the lag. First, training data cutoffs: rebrands, new product lines, and market expansions that happened after a model's cutoff simply aren't in the training set. A model can sometimes close the gap through web search, but only if it decides the query needs it or the user explicitly enables search — left on autopilot, many models default to stale training data.

Second, inconsistent sources. If a company's website reflects its current positioning while social profiles, third-party listicles, old interviews, and review sites carry the old one, the LLM must pick a version to trust. It might not pick yours.

Third, volume. If a company described itself one way for a decade, that's years of reviews and authoritative citations the new positioning must outnumber. Semrush notes the new positioning can be accurate and well documented and still lose the popularity contest.

Four documented cases

Hotjar. Contentsquare acquired the behavioral analytics tool in 2021 and merged with it by July 2025. Yet when Semrush asked ChatGPT for the best heatmapping software, Hotjar ranked second as a standalone product. Hotjar's site redirects to Contentsquare, but the chatbot's answer contained no clickable link, so users never reach the acquirer's ecosystem. Contentsquare ran press releases, blog posts, support documentation, YouTube content, and a Hotjar landing page — and ChatGPT still omitted the acquirer more than a year post-merger. That's an awareness problem for Contentsquare, which paid for Hotjar's name recognition and customer base.

Blue Apron. The meal delivery service dropped its subscription-only model in August 2025, adding pay-as-you-go. It announced the change on Good Morning America, earned coverage from ABC News, Yahoo, and CNET, ran commercials, gave press interviews, and rewrote its homepage. ChatGPT still didn't mention the subscription-free option — which risks alienating exactly the customers the repositioning targeted.

Duolingo. The app added math in 2022, music in 2023, and chess in 2025, backed by Duocon announcements, blog posts, app store description updates, press releases, and a homepage rewrite. Asked for the best apps to learn something new, ChatGPT recommended Duolingo only for languages and pointed to other platforms for math and science.

Peep Laja. The founder of CXL, Speero, and Wynter moved from CRO to business strategy around 2016. He has published a B2B strategy newsletter and weekly podcast for years and lists Wynter on LinkedIn and his personal site. Asked who to follow for CRO, ChatGPT still put Laja first — a decade-old reputation outweighing a sustained, consistent repositioning effort.

The pattern spans enterprise products and solo consultants, and it shows that even well-funded visibility campaigns don't guarantee an LLM catches up quickly.

How to check and fix your AI perception

Semrush recommends auditing what LLMs say before fixing it. Its Perception report shows the features AI platforms associate with a brand, for both branded and non-branded questions, and surfaces the underlying answers for review.

The remediation playbook runs across PR, content, and monitoring:

• Announce changes with PR. Publish the press release on your own site for a canonical version, distribute via wire service, and ask trade publications to cover the pivot. State the new category, audience, and description explicitly, with a clear timeline of what changed and when.
• Update old content. Audit pages for legacy category and product names, redirect outdated pages, and add clarifying statements to high-traffic legacy pages. Use the Backlinks tool to find external sites with outdated anchor text and request corrections, prioritizing the highest-authority links.
• Keep messaging consistent site-wide. Align homepage, about page, product pages, footer, meta descriptions, and title tags. Update internal links and anchor text so old terminology stops reinforcing the previous positioning.
• Keep business listings current. Verify Google Business Profile, LinkedIn, G2, Crunchbase, Wikipedia, and industry directories — LLMs reference these external sources frequently. Listing Management syncs the data from one dashboard.
• Structure pages for LLM extraction. Keep sections self-contained, since AI retrieves excerpts rather than full pages. Use headings that match real user questions and answer them in the copy that follows — naming the category, stating the current audience, and dating the shift.
• Monitor what LLMs say. The AI Visibility Toolkit tracks mentions, citations, and topics where competitors appear and you don't. Semrush frames missed topic opportunities as equally damaging as inaccurate information: if you're absent from a relevant prompt, you're absent from the conversation.

Semrush stresses that PR is a starting point, not the whole strategy — a fresh announcement still competes against years of older content and third-party sources reinforcing the previous story.

For teams that have repositioned recently, the next step is a baseline audit of branded and category prompts across major chatbots, then recurring monitoring to catch drift before an outdated answer hardens into the model's default version of your brand.