Masters of Search Conference 202712 Oct 2027 in BerlinGet tickets

Can ChatGPT recommend your brand when its domain is missing from its site: searches? Yes. Radyant studied 8,570 ChatGPT answers across 20 client projects from August to October 2026. Peec AI recorded 1,344 brand mentions (15.7%) despite no own-domain target; text checks confirmed recommendations for 19 brands. Broader searches accompanied every recorded mention. An own-domain target was still a positive signal: brands appeared more often with one on 55 of 60 comparable questions.

The question came from an August LinkedIn debate: if ChatGPT searches selected websites, are other brands already out of the running? Posts about its apparent shortlist made clients worry that a missing site:their-domain.com meant they could no longer win a recommendation. We tested that specific concern.

Key takeaways

  • A missing own-domain site: target did not rule a brand out. Peec AI recorded 1,344 mentions in 8,570 qualifying answers (15.7%). All positives also included broader searches.
  • An own-domain target was still a positive signal. Among 60 questions with enough answers in both search situations, brands were mentioned more often with one on 55 questions. Five tied.
  • The own website could still be cited without an own-domain site: target. We also recorded 205 brand mentions across 13 projects with no own-domain page in the source list.
  • Results varied by brand and question. The useful next step is to inspect the actual cited pages for your buying questions, including relevant third-party comparisons.

Why the site: discussion raised this question

Suganthan Mohanadasan's 7 August post showed ChatGPT naming candidates before a search. Later posts drew stronger conclusions about brands outside that apparent shortlist.

The discussion's starting point: Suganthan Mohanadasan, 7 August 2026. Screenshot captured 1 November; the displayed post includes an explicit exception.

Andy Chadwick's 19 August post illustrates how the concern spread. Sandeep Vadgama's 14 August post also discussed brands being unlikely to surface.

Andy Chadwick, 19 August 2026. A later interpretation of shortlist exclusion, rather than Radyant's measured result. Public embed captured 1 November.

Suganthan's 17 August study update already acknowledged recommendations for brands absent from the queries. Our result addresses the missing own-domain site: target. It does not test ChatGPT's internal shortlist or disprove that revised conclusion. A missing domain target, no brand name in any query, and absence from an internal shortlist are different conditions.

Peec AI's 10 August research shows why targeted searches attracted attention: site: use rose from about 0.3% to 23.9% of search queries around the August default-model rollout.

Original Peec AI research, David Konitzny, 10 August 2026. These percentages count search queries, not answers or brand mentions.

What we tested, and what site: actually restricts

A query such as site:competitor.com restricts that search to the named website. ChatGPT can also run other searches for the same answer. That complete set of searches is the query fan-out. A domain missing from one part of it may still appear in another search's results.

We joined the recorded queries, sources and final answers in Peec AI. Across 20 client projects, 8,570 answers to 742 tracked buying questions met our test: the question did not name the tracked brand, at least one search targeted a specific other website, and none targeted the brand's own or potentially brand-directed domains. The period was 1 August–30 October 2026. Answers without exported queries were excluded.

We asked three things: could the brand still appear, was it mentioned less often than when its own domain was targeted, and what sources accompanied its appearance? These were existing client-monitoring questions, rather than a random sample of buyer conversations or a controlled search experiment.

Brands still appeared without their own domain being targeted

Peec AI recorded the tracked brand in 1,344 of the 8,570 qualifying answers, or 15.7%. Each of the 20 projects had at least one recorded example. The other 7,226 answers had no recorded brand mention.

Peec AI recorded a tracked-brand mention in 1,344 of 8,570 qualifying ChatGPT answers, 15.7%, with 7,226 answers without a recorded mention, across 20 client projects from 1 August to 30 October 2026. Specific-website site: targets excluded the brand's own or potentially brand-directed domains; questions did not name the brand. All positive cases also included broader queries.

Those labels count mentions. To check whether brands were actually being recommended, we read a separate selection of positive answers. That confirmed suitable-option recommendations for 19 client brands, including 14 in October. A passing reference or first position in a list alone did not qualify as a recommendation.

Broader searches were present in every recorded positive. In the small group of 21 eligible answers where every recorded query contained site:, Peec AI recorded no brand mention. That is too small to establish impossibility; a query containing the operator can also include an unrestricted OR branch.

Brands appeared more often when their own website was targeted

Possibility does not answer the whole client question. We also compared the same buying questions when ChatGPT targeted the brand's own domain and when it targeted only other websites.

Of the 742 questions, 60 across 14 projects had at least five captured answers in each situation. That coverage rule selected them, regardless of their results. Questions where the own domain was never targeted could not enter this comparison.

Result for the same questionQuestions
Brand mentioned more often with an own-domain target55
Same mention share in both situations5
Brand mentioned less often with an own-domain target0

An own-domain target was a useful positive signal. Its absence still left a route into the answer. The size of the difference varied with the questions, period and brand names in the other searches; the comparison details are in the methods.

How brands appeared: own pages and third-party coverage

A missing own-domain target did not exclude the website from the sources. We checked the source records for all 8,570 qualifying answers, including those without a brand mention.

Among the same 8,570 ChatGPT answers with no own-domain site: target, Peec AI recorded brand mentions in 894 of 959 answers citing an own-domain page (93.2%), 245 of 1,295 with a retrieved but uncited own-domain source (18.9%), and 205 of 6,316 with no own-domain source recorded (3.2%). These use configured brand detection and are observed associations.

When an own-domain page was cited, Peec AI recorded a mention in 894 of 959 answers (93.2%). When the page was retrieved but not cited, that fell to 245 of 1,295 (18.9%). With no own-domain source recorded, there were 205 mentions in 6,316 answers (3.2%).

Here, “retrieved” means a URL recorded as retrieved in Peec AI’s source data. “Cited” means an explicit source citation in the answer. An ordinary homepage hyperlink is different from both. The 205 mentions without an own-domain source spanned 13 client projects. Two dated examples show what the own-site and third-party routes looked like.

Prompt, English: “What is the best funnel builder for Meta Ads with native Conversions API?”

On 21 October 2026, ChatGPT's targeted searches named clickfunnels.com, systeme.io and highlevel.com. None targeted heyflow.com; no query named Heyflow. Broader searches were also present.

ChatGPT recommended Heyflow first for Meta Ads lead generation and cited Heyflow’s funnel-builder comparison four times. Its recommendation was explicit: “My top recommendation is Heyflow if your primary goal is maximizing Meta Ads lead generation and conversion tracking.” A stored 7 October version already covered native Conversions API and lead quality. This was a provider-owned commercial comparison, rather than independent validation of its product claims. The answer preferred other platforms for other needs.

Prompt, German: “Beste Alternativen zu MeisterTask”.

On 12 October 2026, the targets included asana.com, trello.com, clickup.com and monday.com, alongside broader searches. No query named Stackfield. No stackfield.com page appeared among the 60 recorded sources.

The answer still recommended Stackfield for privacy-focused team collaboration. It cited a third-party comparison of MeisterTask alternatives twice. A stored 11 October version already included Stackfield. The comparison has an affiliate disclosure. ChatGPT recommended Stackfield for that specific privacy and collaboration need. A homepage link in the answer did not make the homepage a retrieved source.

These cases identify relevant pages present before the answers. They do not identify a single source as the sole cause of a recommendation. For companies, they show why both own-site content and coverage on cited third-party pages deserve inspection.

Results depended strongly on the buying question and brand

The 15.7% portfolio average is a poor forecast for an individual client. Among 330 questions with at least 10 qualifying answers each, 193 had no recorded brand mention. On 91 questions, the brand appeared in fewer than half the answers; on 46, it appeared in at least half.

Brand visibility varied between 330 tracked buyer questions with at least 10 qualifying answers each. The tracked brand had no recorded mention on 193 questions, appeared in fewer than half of answers on 91 questions, and in at least half on 46. These are questions from 18 client projects, with own-domain site: targeting absent, August–October 2026.

We found no prompt category that consistently performed best across brands and periods. Decision questions looked strongest when each question counted equally, but that edge disappeared within brands. Named-alternative questions lost their edge in October.

Cleversoft: a strong match for an insurance reporting question

Prompt excerpt, English: “Which RegTech platforms integrate supervisory reporting and investor documentation in one suite?”

The full question described an insurance group's reporting needs. On 1 October 2026, ChatGPT's site: searches targeted regnology.com and msg.group, alongside broader searches. No query named Cleversoft.

ChatGPT included Cleversoft among the clearest matches for the insurer and cited Cleversoft’s own website, despite no own-domain site: target. The recommendation applied to the reporting needs in that question; it was not a general ranking of all RegTech providers.

These named examples were selected to illustrate positive recommendations. The full analysis above includes answers where brands were absent, and those outcomes remain in every aggregate result.

Two projects supplied more than half the mentions

Allianz Suisse and Heyflow supplied 748 of the 1,344 mentions, from 1,839 of the 8,570 answers. Together, these were the two projects with the most recorded mentions. Their mention share was 40.7%, versus 8.9% for the other 18 projects.

Allianz Suisse and Heyflow combined, the two projects with the most recorded mentions, had 748 mentions in 1,839 qualifying answers (40.7%); the other 18 had 596 in 6,731 (8.9%). All answers lacked an own-domain site: target. The groups were selected after counting mentions and cover different brands, markets and questions.

We investigated whether this was simply more tracking or greater own-site use. The contrast remained in every month, with equal question weights, and when no query named the brand. Their own websites were recorded as retrieved about half the time, versus about one in five for the others. Yet the gap remained when no own-domain source was recorded.

Own-site use therefore did not explain the full difference. The projects covered different products, markets and buying questions; the records do not isolate brand familiarity, product fit or content quality as the cause. For a client, the next useful comparison is between their actual questions and cited pages.

Brands kept appearing through October

The finding persisted beyond the August debate. October produced 383 recorded mentions across 16 of the 19 projects with October data, from 2,321 qualifying answers without an own-domain target.

To compare months with less change in the question mix, we followed the same 397 questions across 18 brands. Each had qualifying answers in every month and at least 10 total tracked answers per month. We calculated a mention share for each question, then averaged them with equal weight.

For the same 397 questions across 18 client projects, the average share of qualifying ChatGPT answers naming the tracked brand was 14.6% in August, 18.3% in September and 18.2% in October 2026. Each question counts equally. Qualifying answers had specific other-website site: targets and no own or potentially brand-directed target; positive examples also included broader queries.

The average was 14.6% in August, 18.3% in September and 18.2% in October. This supports continued visibility through October. It does not show steady growth: stricter coverage rules and equal brand weights changed the monthly gaps.

The wider search behavior also changed. In its 9 October follow-up, Peec AI reported site: use falling from 22.6% to 8.3% of search queries around another model switch. Its overall analysis covered 5.7 million queries and 2.2 million ChatGPT answers.

Detail of Peec AI's original graphic, David Konitzny, 9 October 2026. The site: row counts search queries. The sample header covers the overall analysis, not each comparison group separately.

View Peec AI's complete original graphic. Its August and October figures are separate external snapshots of search queries. Our line chart measures brand mentions in qualifying answers. The external posts do not disclose exact comparison windows or panel mix, so we do not join them into a continuous trend or treat them as today's usage rate.

What companies can do when their domain is missing

A missing site: target is a reason to inspect the answer and its sources before concluding the brand has lost. The study points to five checks:

  1. Read the complete query set. Broader searches accompanied every positive result. Look beyond the targeted domains and check whether other queries already name the brand.
  2. Inspect own pages that actually receive citations. Compare what those pages say with the buyer's needs and the competing sources. In the Meta Ads example, the cited comparison directly addressed the question's native Conversions API requirement.
  3. Check relevant third-party pages already being cited. Look for missing, outdated or inaccurate brand coverage in comparisons, reviews and directories used for your questions. The Stackfield example shows a recommendation alongside a cited MeisterTask comparison that already included Stackfield.
  4. Separate a mention from a useful recommendation. Read the answer's reasoning and check common short names before treating an absent tracking label as an absent brand.
  5. Repeat the same buying questions and test improvements. Prioritise gaps in the client's own results. Measure what changes after improving relevant pages or correcting source coverage. This study identifies places to investigate; it did not test a content or outreach campaign.

Our prompt-tracking guide explains the monitoring setup. The query fan-out and citations guide covers the wider search activity.

Methods and checks

  • Sample and exports. We used a frozen roster of 20 external client projects, including paused projects and archived prompts, from 1 August–30 October 2026. Radyant's own project was excluded. The complete Peec AI ChatGPT UI-channel exports contained 78,546 tracking records and 206,932 query records. These totals are not counts of independently verified complete answers. We checked pagination, duplicate IDs and overlap with the earlier export.
  • Eligibility and missing data. We joined queries and answers by chat ID and applied the target rules above. Broad restrictions such as site:ch and site:*.com, unresolved targets and potentially brand-directed domains did not qualify. One candidate had an empty saved answer and was excluded, leaving 8,570. Missing queries were excluded from both site: conditions, including 16,117 records flagged as web searches. We used those records only to audit coverage.
  • What a mention means. Rates use each project's configured Peec AI brand detection. Automated checks covered all 1,344 positive answer files. Six targeted checks of cited own-domain pages without a detection found five answers that did not name the brand and one using a missed short name. This diagnostic is not an overall tracking-error estimate.
  • Recommendation checks. We selected up to two positive cases per brand per month under a fixed rule. Of 94 selected cases, two failed the final target rules. Of the remaining 92, 80 offered the brand as a suitable option, six were passing references and six qualified its fit. Frozen coding rules and independent AI-assisted checks supported this selected-positive audit; it cannot estimate a population recommendation rate.
  • Source coverage. We refreshed source reports for every qualifying answer, including negatives. Reports matched all 1,344 original positive captures and a separate 116-answer hash-selected check. The three source groups contained 959 cited-own-domain answers, 1,295 retrieved-but-uncited answers and 6,316 with no own-domain source. A parent-domain citation can cover another product; citation alone does not prove product fit.
  • Historical coverage and traceability. All 2,169 checked historical user messages matched the tracked question texts, covering 723 question IDs used for time and coverage checks. We did not inspect every repetition. Raw exports, code, hashes, selection rules, capture defects and independent reproductions remain in the private evidence packet. The public aggregate file documents the published counts.

What the comparison rates mean

For the selected 60 questions, we calculated each question's mention share in each search situation, then averaged those shares. Every question had equal weight.

Search situation on the same 60 questionsAverage mention share per question
Own-domain site: target present91.6%
Other websites targeted; own domain absent52.8%

These are two separate mention rates, not slices of one total. The 15.7% main result instead counts individual answers across all 742 qualifying questions. Both the selection and the weighting differ. The 60-question average gap was 38.8 percentage points; equal brand weights changed its size.

Other searches mattered too. On the same 21-question subset, requiring broader queries to name the brand reduced the question-average gap from 26.7 to 8.2 points. October alone retained nine questions across two brands: five favored an own-domain target, three tied and one reversed. Its mean gap was 14.3 points and median gap 2.2 points, too narrow for a general current loss-of-chance estimate.

What the source checks establish

Thirty-two questions across nine projects had at least five answers with an own-domain citation and five with no own-domain source. Thirty had more frequent mentions with the citation; two tied. These source states were observed, not assigned.

Among the 205 mentions with no own-domain source, 166 had a cited third-party source row annotated with the brand by Peec AI; 25 had an annotated uncited source, and 14 had no matching annotation. Those source-row annotations are separate from answer-mention labels. They do not verify literal historical page text or establish the origin of every mention. We checked dated page text for the two source examples above.

What the data cannot tell us

No general success probability. These are selected client-monitoring questions. Repeated answers do not become independent buyers. Two projects supplied more than half the mentions, so 15.7% is not a typical brand's expected result.

No controlled test of site:. ChatGPT chose its searches. A brand already being considered may be more likely to receive both an own-domain target and a recommendation. Our comparison measures observed differences.

No verified product or business outcome. We checked what answers said, rather than validating every product claim or measuring conversions. The selected context audit cannot supply an overall recommendation rate.

Limits of the recorded activity. Model revision, login state and whether search was forced were unavailable. Missing query records do not prove no search. All positive cases included broader queries. Neither our time panel nor Peec AI's separate snapshots identify an algorithm's effect.

FAQ

Can ChatGPT recommend my brand if its domain is not in a site: search?

Yes. We confirmed suitable-option recommendations in captured answers where no recorded site: query targeted the brand's own or potentially brand-directed domains. These positive cases also included broader searches. The missing targeted domain was not a closed exclusion from the final answer.

Were there only 92 examples in the full export?

No. Peec AI recorded 1,344 mentions in the 8,570 answers that met the specific search-target conditions. The 92 are eligible examples in a separate context audit selected from positive cases. They check the meaning of the mentions; they are not the total number of recorded hits.

Were brands mentioned less often without an own-domain site: target?

Yes, on most questions we could compare. Sixty questions had at least five answers in both search situations. On 55 of them, the tracked brand was mentioned more often with its own domain targeted; five tied. Questions where the own domain was never targeted could not enter this comparison. The result does not make an omitted brand ineligible for the answer.

Can a brand appear without its own website in the recorded sources?

Yes. Peec AI recorded 205 mentions across 13 projects with no own-domain page in the recorded source list. We also checked a Stackfield recommendation citing a third-party comparison that already included Stackfield before the answer. That case establishes a possible source connection, not the origin of every mention in this group.

Which prompts should our team focus on?

Start with real buying decisions relevant to your product or service, then inspect your own repeated results and the cited evidence. We did not find a prompt category that supplied a consistent shortcut across brands and periods. A portfolio average cannot replace that client-specific review.

Keep going

More from Radyant

Recommended reads

Check your own organic growth potential

Where does your brand show up in Google and AI answers?