"We're not showing up in ChatGPT" is the complaint. There is rarely one clean cause, and no published study measures how often each of these causes is responsible. Below are five places to investigate when an answer is missing or wrong, ordered by how easy each is to check and how much an error there would matter to customers, with a way to test each one yourself before assuming it's the culprit.
1. Weak or unclaimed directory presence
In Yext's API-derived sample of 6.8 million citations, OpenAI model output drew 48.73% from controllable third-party listings.1 That makes relevant directory accuracy one useful audit target. Compare it with the sources shown across repeated ChatGPT checks.
Directory accuracy sits near the top of this checklist because it is quick to verify and correct, not because the cited research proves it gates ChatGPT visibility. Treat it as one source-coverage hypothesis among several.
Search your own business on Yelp, Bing Places, and Facebook
Look for three things on each: is the listing claimed, is the category correct, and are the facts current? Yext's API-derived sample found Gemini model output drew 52.15% from first-party websites and OpenAI model output drew 48.73% from controllable third-party listings.1 Check both the website and relevant directories, then compare them with the sources actually shown in repeated answers.
What fixing it looks like: claim every major directory listing that isn't already claimed, correct the category and business details, add photos, and make sure the description says something specific about what the business does — not a generic filler sentence. Claiming the listing is a one-time task; keeping it accurate is not. Services, hours, links, and other details can drift out of date on their own even when the name, address, and category never change — recheck periodically, not only after a major business change.
2. Reputation is weaker than the visible alternatives
In SOCi's 2026 Local Visibility Index, recommended locations averaged 4.3 stars on ChatGPT, 4.2 on Perplexity and 3.9 on Gemini.2 That is context, not a cutoff. Judge the business against the actual alternatives surfaced in its market, alongside review volume, recency, response quality, and source consistency.
Review quality is a separate check from source coverage. Compare ratings and recent feedback with the businesses that actually appear, but do not infer a fixed cutoff or a sequence in which one factor gates another.
Pull your current rating on each major platform, separately
Google, Yelp, and Facebook ratings can differ for the same business. Check all relevant sources and compare them with the businesses customers are actually offered.
What fixing it looks like: resolve recurring service issues, request reviews consistently and without incentives, and respond thoughtfully to recent feedback. BrightLocal's 2026 survey found 74% of consumers trust only reviews from the last three months and 32% only the last two weeks.3
3. Crawler access blocked in robots.txt
Worth checking early because an explicit denial is a binary retrieval block rather than a ranking disadvantage. If robots.txt disallows OAI-SearchBot, Claude-SearchBot, or PerplexityBot, that specific search crawler cannot retrieve the affected pages. Training crawlers are a separate choice and should not be confused with these search-facing agents.
Some site builders and security plugins add broad crawler blocks by default, sometimes years before "AI visibility" was a consideration, and the block may never have been revisited.
Check intended access, robots rules and real fetch results
Have your website maintainer inspect which public pages you intend these crawlers to reach, the effective robots.txt rules for OAI-SearchBot, Claude-SearchBot and PerplexityBot, and any actual fetch failures in server or CDN logs. A robots.txt that permits a crawler does not guarantee the page can be fetched: firewall, bot-protection or login rules can still block it. Keep private routes and any training-crawler preferences as they are, then retest one specific public page.
What fixing it looks like: removing the disallow rule for the specific crawlers, or adding explicit Allow rules if a broader block needs to stay in place for other reasons. The change itself can be small, but it may need a developer or site-admin access, and a broad rule can affect other crawlers, so retest a specific public page afterwards.
4. Content that's vague instead of specific and citable
Even with directory presence, an acceptable rating, and no crawler block, a business can still underperform if its own content gives an AI model nothing concrete to cite. The 2024 GEO (Generative Engine Optimization) paper reports gains of up to 40% more visibility on its own benchmark, varying by domain.4 "Award-winning service you can trust" is not citable — there's no fact in it. "Family-owned since 2003, specializing in same-day emergency plumbing repairs" is citable — an AI model can quote or paraphrase a specific claim, and a vague one gives it nothing to work with.
Content specificity affects whether a source contains a direct, verifiable answer worth extracting. It works alongside source coverage, comparative reputation, crawler access, and entity consistency rather than after a simplistic sequence of pass/fail gates.
Read your own homepage's first paragraph as if you were a stranger
Count the adjectives ("best," "trusted," "leading," "premier") against the concrete facts (years in business, specific services, specific location, specific credentials or numbers). If the adjectives outnumber the facts, that paragraph is not doing the job an AI model needs it to do.
What fixing it looks like: rewriting the top of the homepage — and ideally key service pages — to lead with one or two specific, verifiable claims instead of general praise. This doesn't require new copywriting skill so much as a willingness to cut the adjectives and replace them with facts that are already true about the business.
5. Genuinely low overall web footprint
The most straightforward to diagnose: some businesses are simply too new, or have too little presence anywhere online — no reviews yet, a bare-bones website, no directory listings — for any AI model to have signal to draw on. This isn't a "fix one thing" problem the way the other four are; it's closer to "there isn't enough material yet for any engine to cite," and the answer is building basic presence across the board rather than optimizing one specific gap.
It comes last here because it is a broader project rather than a single check. It is most relevant to genuinely new businesses, or ones that have deliberately kept a minimal web presence.
Search the business name plus category across Google, Yelp, and Bing
If almost nothing comes back anywhere — no reviews, no directory listings, a thin or missing website — the issue isn't a specific broken signal, it's an absence of signal altogether. That's a different, larger project than fixing one gap.
What fixing it looks like: building the basics: accurate directory listings, genuine reviews, a crawlable site and specific content. A low web footprint is addressed by the first four checks together, not by a separate fifth action.
"Most businesses assume the cause is unique to them. Working through five concrete checks (directories, reviews, crawler access, content, footprint) replaces guessing with evidence."
Notes and sources
1 Yext, "AI Citations, User Locations, & Query Context," Yext Research, Oct 9, 2025. 6.8 million citations from ~1.6 million questions across OpenAI, Gemini and Perplexity APIs, July-August 2025, four industries. In its API-derived sample, OpenAI model output drew 48.73% from controllable third-party listings and Gemini drew 52.15% from first-party websites. API citations can differ from direct user applications.
2 SOCi, "2026 Local Visibility Index," January 2026. Dataset: 2,751 multi-location brands, about 350,000 US locations. Benchmarks for ChatGPT, Perplexity and Gemini: share of locations recommended 1.2%, 7.4% and 11.0% (Google 3-Pack: 35.9%); average rating of recommended locations 4.3, 4.2 and 3.9 stars; business-profile accuracy 68.3%, 68.0% and 100.0%, measured against Google Maps data.
3 BrightLocal, "Local Consumer Review Survey" 2026. Sample: n=1,002 US consumers. Findings cited: 74% of consumers trust only reviews from the last 3 months; 32% trust only the last 2 weeks.
4 Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande, "GEO: Generative Engine Optimization," submitted November 2023, published KDD 2024. Finding: "GEO can boost visibility by up to 40% in generative engine responses" on the paper's GEO-bench benchmark; the effect varies by domain.
RaveHQ Insights share our views, based on our research and the sources cited. They are general information, not legal, financial or professional advice; check what applies to your business before acting.