Want ChatGPT to name your business when someone asks about your field? Three things we'd check first: state the facts a buyer decides on (price, who it's for, who provides it) in plain text near the top of the page, make sure AI crawlers can reach that page, and get mentioned on the sites these tools already read. Below is what we saw in our own tests, not a claim about how the algorithm works, since nobody outside OpenAI knows that.
One reason this is worth doing: when we scored real dental-implant pages in September 2026, about 87% stated two or fewer of five buyer facts in their opening. Most pages simply don't say enough for an AI to repeat. That's the gap the steps below close.
6 things we'd fix first for ChatGPT visibility
1. Put your buyer's facts in the first ~100 words
This is where most sites fall short. AI answers frequently reuse extractable facts from retrieved pages, and analyses of Google's AI Overviews find the opening carries a big share of what gets quoted (one study measured ~38% of cited passages in the first 100 words). If your price, your named provider, who the service is for, or your guarantee sits only lower down, it's not in the part we measured.
What to do: open the page with a plain answer to the question a buyer actually asks, and name the real facts right there. Not "we offer competitive pricing," but the actual number, the actual credential, the actual detail. In our scan, price showed up in the first 100 words on only ~3% of pages: the fact buyers most want is the one most often missing.
2. Get mentioned on sources these tools already trust
The old backlink instinct only half applies. Published citation studies suggest that mentions (your name showing up, consistently, on sources these tools already read) track with getting cited. Same name, same core facts across your site, your profiles, and third-party write-ups. Contradictory facts give an AI nothing solid to repeat. Earn mentions in trade press, roundups, interviews, credible directories. Think digital PR, not link volume.
3. Be where these tools look: Reddit, YouTube, "best-of" lists, directories
AI answers are assembled from retrieved sources, so be present on the sources that get retrieved. Community platforms like Reddit, YouTube (transcripts are text an AI can pull), "best X" comparison lists in your niche, and credible directories all put your name where these tools read. One caveat: which surfaces matter shifts by engine and by month, so check which sources ChatGPT actually cites for your questions instead of assuming a universal list.
4. Keep pages current
Across this topic, the pages that show up are mostly dated to the current year and updated recently; stale pages are largely absent. Put real dates on pages and actually refresh the facts, not just the date. Revisit cornerstone pages on a schedule: if your price changed, the page should say so. Freshness is one of the few things you fully control.
5. Structure the page so a machine can read it
Lead sections with the answer, then the detail. Use question-shaped headings, clean H2/H3 hierarchy, short paragraphs, and lists where they fit. Schema markup (a type that genuinely fits the page) is useful hygiene that helps a machine read what the page is, not a shortcut, and it won't rescue a page that states nothing.
6. Make sure an AI can actually reach the page
You can do everything above and still be invisible if you've blocked the door. Check robots.txt for the AI user-agents you want to allow: for ChatGPT Search the one that matters is OAI-SearchBot (GPTBot is about training; ChatGPT-User is live user-triggered fetches). Check any Cloudflare/CDN "AI crawlers" toggle. Submit your site to Bing Webmaster Tools, since ChatGPT Search draws on providers that include Bing. (llms.txt is an early, unproven convention. Google has said its Search doesn't use it, so treat it as a low-cost experiment, not a lever.)
Before (0 of 3 core facts up top): "We are a patient-focused implant practice committed to excellence and personalized care in a comfortable environment."
After (3 of 3 core facts, still readable): "Dental implants in Austin start at $1,800 per implant, placed by a board-certified surgeon with 12 years' experience. Good candidates have healthy gums and enough jawbone; we confirm with a scan at the first visit."
The "after" isn't keyword stuffing: it answers what a buyer asks and states the facts plainly, so a person reads it well and a machine has something concrete to reuse. (These names and numbers are illustrative.) Do the same with your own real numbers.
A simple way to check yourself
Two different problems, two checks.
Step 1. Can an AI reach your page? Confirm OAI-SearchBot is allowed and the page renders as crawlable text.
Our AI crawler check reads your domain and shows which AI crawlers you allow or block, plus sitemap and llms.txt signals. A green result means you're visible over HTTP, not that a crawler definitely reaches the page — a CDN or firewall can still block it in ways an external check can't see.
Step 2. Does the page actually say anything useful? Read only the first ~100 words and score five decision-facts (price, financing, provider, candidacy, and options), one point each. The 0 to 5 total is your opening's readiness, not a prediction of anything. You can do this by hand or paste the opening into ChatGPT with a short scoring prompt; the full method is on our 100-Word AI Readiness Test.
Then keep a short log: same questions, on a schedule; record engine, date, region, the answer, and the sources named. Results vary between runs, so track a pattern over time, not a single verdict.
ChatGPT vs Google: how the playbooks differ
Same foundation (good content, real authority, a crawlable site), different emphasis. The biggest difference is what you measure: on Google you track a fairly stable position; on ChatGPT you track how often you're cited across queries and repeated runs.
| Google (classic) | ChatGPT / AI answers | |
|---|---|---|
| What you track | a relatively stable position | citation/mention likelihood across queries + repeat runs |
| What you're after | a ranked position | being a cited source |
| Opening facts | useful for snippets | what our opening scan checks |
| Freshness | matters | appears to matter (recency tracked in some analyses) |
| Structure | helps rankings | helps extraction |
| Reachable via | Google index | third-party search providers + OAI-SearchBot (per OpenAI) |
| Result stability | fairly stable | varies run to run |
How we know this: method and limits
We ran our own studies instead of repeating other people's numbers. Read these as observations in specific fields, on dated runs, reported as association, not causation.
- Opening scan (dental implants, 2026-09-05): 144 pages across 17 US metros, scored 0 to 5 on whether the opening states five buyer facts. Mean 1.64; about 87% scored two or fewer; price appeared in ~3% (4 of 144). Each page was scored in two passes and disagreements resolved by a fixed rule (99% agreement between passes).
- Wider index (2026-09-06): the same pattern held across 852 page/location records in seven unrelated industries, about 84% scored two or fewer. These are page/location records, not verified independent businesses, and each industry used its own fact set.
- Which pages got cited: on a proxy AI-search setup (not the standard ChatGPT app), run three times, pages that stated 3 to 4 facts were cited in at least two of three passes about 1.8 times as often as pages stating 0 to 2 facts (36.8% vs 20.4%). On a looser definition it's about 1.4 times. This is one slice, not a rule for every query. Across the three passes the specific sources named kept shifting, though the fact-rich vs fact-poor gap held, which is why we suggest tracking a distribution across runs rather than a single position.
- A regression on 174 clinics found page-fact richness was associated with being cited (odds ratio ~2.0, 95% CI 1.4 to 3.0, p<0.001). Across 7,389 reviews from the clinics in this analysis, the review metrics we tested (count and content) did not separate who got cited; page-fact richness did. "Not significant" means our data did not show an effect, not that reviews don't matter.
- Cost questions: of 150 AI Overview queries, 81 returned an answer; among those, all 50 cost/price queries showed an AI Overview, and the pages cited for cost usually stated a price.
What we don't claim
We don't know how ChatGPT decides; everything here is "what we observed" or "what the platform states." No one can promise "#1 in ChatGPT": there's no stable position to defend, and results shift run to run. Our numbers show what tracked with getting cited, in specific fields, on dated runs, strong enough to act on but not a guarantee.
Full methodology and data: 100-Word AI Readiness Test - Which pages AI cited
Sources. Our own figures link to /research; each external claim is scoped to exactly what its source measured. Position of cited passages within a page, on Google AI Overviews (not ChatGPT): Surfer, ~38% in the first 100 words. ChatGPT Search draws on third-party providers and needs OAI-SearchBot allowed: OpenAI Help, "ChatGPT search". OpenAI's crawlers are separate — GPTBot (training), OAI-SearchBot (search), ChatGPT-User (user-triggered fetches): OpenAI, "Overview of OpenAI crawlers". Google Search does not use llms.txt: Google Search Central, AI-optimization guidance. Reddit's citation share in ChatGPT dropped in mid-2026 (reported provisional): Promptwatch. AI-cited URLs tended to be more recent than top-ranking pages in one analysis of AI log hits: Seer Interactive.