How to show up in ChatGPT and AI search

Most advice about appearing in AI answers is written by people who have never measured one. So in September 2026 we measured 48 of them, against eight companies you have heard of, and the result was not what the advice predicts.

What we measured, exactly

On 26 September 2026 we put six buyer-intent questions to Google Gemini with web search grounding switched on, once per company, for eight monitoring and observability vendors. That is 48 answers, every one of them drawn from a live search rather than the model's memory — we checked the grounding metadata on each, and all 48 came back grounded.

The questions were the ones a buyer actually types, not brand names:

• best uptime monitoring tool for small teams
• cheaper alternative to Datadog
• best tool for centralized log management
• how to get alerted when my website goes down
• affordable observability platform for startups
• best status page tool

For each answer we recorded two different things: whether the company was named in the text, and whether its own domain was cited as a source. Those turn out to be almost unrelated, and the difference is the entire point of this article.

The result

Across all 48 answers: named 19 times (40%), cited 2 times (4%).

Datadog is one of the best-known names in the category. It was named in five of six answers and cited in none of them. Better Stack, same: five mentions, zero citations. Two citations turned up in the whole study — one for New Relic, one for Instatus.

If you have been told that being a recognised brand is enough, this is the counter-example. The engine knows who you are. It is sending the click somewhere else.

Why being named is not being cited

A grounded answer is assembled in two passes. The model writes what it knows — that is where the brand names come from, and they come from training data that may be a year old. Then it runs a search to support what it wrote, and cites whatever it finds.

What it finds is almost never the vendor. It is whoever wrote the comparison post.

Ten domains, and every one of them turned up in the answers for all eight companies. Several are competitors who happen to have written a round-up: Uptrace, UptimeRobot, OpenObserve, OpenStatus. One is Reddit. Not one is a category leader's own marketing site.

This is the mechanic nobody mentions: the citation goes to the page that answers the question, not to the company the answer is about. Your homepage explains what you sell. A listicle titled “12 best uptime monitors compared” answers what was asked. The engine picks the second one, every time.

The checklist that actually moves it

In rough order of effect, based on what the cited pages have in common.

1. Publish the comparison yourself. The pages being cited are comparisons, alternatives and “best X for Y” round-ups. If a competitor's blog is the source for “cheaper alternative to Datadog”, that is a page you could have written, honestly, including where you lose. The engines cite pages that compare; they skip pages that sell.

2. Submit to Bing Webmaster Tools. ChatGPT search and Copilot run on Bing's index, not Google's. It is free, takes ten minutes, and a site absent from Bing is invisible to a large share of AI traffic regardless of its Google rank.

3. Mark up what you are. Organization, Product and FAQPage schema, with the visible text matching the markup. Answer engines parse structured data to decide what a page is about before they decide whether to quote it.

4. Get onto the aggregators that get cited. G2, Capterra, Reddit threads in your category. Capterra and Reddit both appeared in our results. These are citations you do not control but can participate in.

5. Claim your Google Business Profile if you are local. For “web design Langley”-shaped queries, the profile is often the entity the answer is built around.

6. Add an llms.txt. Cheap, not yet widely honoured, and worth doing on the chance that it is. Do not expect it to carry the work — nothing in our data suggests it is being read today.

What is conspicuously missing from that list is keyword density, word count and the rest of the 2015 playbook. None of the cited pages won on those.

What this study does not show

Six questions in one category, one engine, one day. That is enough to show the gap between being named and being cited — the gap is far too wide to be sampling noise — but it is not a ranking of these eight companies, and you should not read it as one.

Three limits worth stating plainly:

• One engine. We used Gemini with Google Search grounding. ChatGPT runs on Bing and will cite a different set of pages. The mechanic is the same; the specific domains will not be.
• One sample each. Ask the same question twice and the answer can differ. Anything resting on a single cell is weak; the pattern across all 48 is not.
• Citation is not traffic. Being cited means a link appeared under the answer. Whether anyone clicked it is a question this method cannot reach.

The two citations we did record are instructive. New Relic was cited for “affordable observability platform for startups” — a query where its own pricing page answers the question directly. Instatus was cited for “best status page tool”, which is the one thing it does. Both won on the query where their own page is the answer, not where they are merely relevant to it. That is the shape of what works.

Measure it rather than assume it

Every number above came from asking the question and reading the answer. You can do the same by hand: open the engine, type what a buyer would type, and write down who is named and which links appear underneath. Six questions takes about twenty minutes and tells you more than any audit.

We built a scanner because doing it by hand every week does not scale, and because a single reading is noise — the same question asked twice can come back differently. It reports the citation rate, the mention rate, who is recommended instead, and the domains those answers cite, which is the outreach list. The first scan is free.

Whatever you use, measure the two numbers separately. A company that is named but never cited has a completely different problem from one that is neither, and the fixes are not the same.

If you are weighing how much of this to invest in, the companion question is whether AI is replacing SEO — the short answer is no, but what changes is worth understanding before you spend anything.