ChatGPT placed Mindo first in 21 of 134 questions in July 2026, up from 4 in the same questions in June

25 and 26 June 2026, ChatGPT

16 of 134 (11.9%)134 spontaneous questions about Mindo’s market, answered by ChatGPT in both cycles using the same standard and judge. The guide had been live since 22 June, so this is not the starting point of the work.

→

13 and 14 July 2026, ChatGPT

31 of 134 (23.1%)the same 134 spontaneous questions, the same standard and judge: 24 questions gained a recommendation and 9 lost one

period

The opening diagnostic took place on 29 May 2026. The before measurement was captured on 25 and 26 June 2026, and the after measurement on 13 and 14 July 2026. The order matters: Mindo’s entire guide was already live by 22 June, so the June measurement is not a snapshot from before the work. The pre-work snapshot is from 29 May, and it uses a different denominator.

company size

Not reported. Mindo’s size is not recorded in its profile or interview, and murmur.marketing does not estimate what it has not measured.

question set

134 spontaneous questions that ChatGPT answered in both cycles. Claude’s equivalent set has 123 questions; Google’s has 167. July’s full set contains 538 spontaneous questions per engine, after excluding the 27 containing the word Mindo. The 29 May diagnostic covers a different set: 63 captures, from 21 questions across 3 engines.

engines

ChatGPT, Claude and Google, in both cycles. Perplexity ran in June but was excluded from July, so every comparison uses only the three engines present in both. The May diagnostic covered 3 engines: Perplexity was blocked by Cloudflare and never ran.

On 29 May 2026, Mindo appeared in none of the diagnostic’s 63 captures

Mindo is a São Paulo motion-design studio, registered as MINDO PUBLICIDADE LTDA under CNPJ 00.319.345/0001-02. It serves businesses in two lines: corporate presentations and animated video. It is a client of murmur.marketing, a Brazilian Generative Engine Optimization operation run by Mateus Gomes, which was hired for the work described here and measures its own results. The figures on this page come from reports from that same operation, and the case is published with Mindo’s permission, recorded on 10 August 2026.

On 29 May 2026, before murmur.marketing had published anything, Mindo was not mentioned once. The diagnostic ran 21 questions from its market across three engines, producing 63 captures, and Mindo appeared in none of them. This was not one engine’s bias: all three agreed, across both business lines.

The cause was the website. It was a React application that assembled the page in the browser without server-side rendering, so an AI crawler that did not run JavaScript saw an empty page. There were six institutional pages and nothing else: no blog, articles, FAQ or schema, and the HTML language was declared incorrectly. It was not blocked: robots.txt allowed everything and a sitemap listed the six pages. There was nothing to read.

That zero is valid only with its own denominator. It counts mentions in 63 captures from May. The scorecard counts recommendations in 134 ChatGPT questions in June and July. They are different instruments and universes, and murmur.marketing does not stitch them into one line.

There were two markets, not one, and the barrier on generic questions was an entity problem

The measurement revealed two markets. For presentations, AI named one group of companies. For animated video, it named another. Only one competitor crossed both. Because Mindo sells both, the work was divided into two territories from the first day.

There were also orphan questions, in which none of the three engines anchored any supplier: sustainability and ESG reports, presentations for corporate events, training video, internal communications and visual communication for pharma. Mindo had a real case in almost all of them. That was where the guide began, published in batches at guia.mindo.com.br. By 22 June the corpus was already entirely live, which is why the 134 June questions, captured three days later, are not a before-work snapshot.

When the measurement showed that the barrier on generic questions was about the entity rather than content, work opened beyond the site: a dossier grounding company facts in open sources and draft listings in a directory and an open database, both dependent on client-side execution. This is long-horizon work and none of it enters the number on this page. One item was refused in writing: the founding year. The company’s claimed date has no third-party open source, and the only anchored date is 1994, the opening of its CNPJ registration rather than the studio’s founding. The recorded recommendation was to omit the field from every listing until an open source appears.

On ChatGPT, 24 questions began recommending Mindo and 9 stopped

In the remeasurement on 13 and 14 July 2026, ChatGPT recommended Mindo in 31 of 134 spontaneous questions, compared with 16 in the same 134 on 25 and 26 June. That is an 11.2 percentage-point difference. The question-by-question account explains it: 24 questions that had not brought up Mindo began to do so, 9 moved in the other direction, 7 already brought it up and continued to do so, and 94 did not bring it up in either cycle.

The new questions included generic ones such as best animation studios. That is where the owner wants to be, and where the barrier was highest: it is the question a buyer asks before any company is in mind. In those answers, ChatGPT began describing Mindo as an option for original 2D animated video and corporate presentations. That is what was measured: it began to be recommended in those questions. List position was also measured, and it is the hardest number in this case: in the same 134 questions, Mindo was the first answer in 4 in June and 21 in July, with 20 questions entering and 3 leaving. The count follows the standard stated here: first-listed recommendation, spontaneous questions, ChatGPT.

In 104 of 1,695 audited responses the guide influenced the text, and in 90% AI gave credit

There is a legitimate fear in publishing content today: that it feeds AI for free. murmur.marketing investigated this response by response, with auditors reading each one and a second adversarial-review pass that downgraded 9 strong cases before the figure was published. Of the 1,695 responses in the 13 and 14 July 2026 cycle, 104 (6.1%) show real influence from the guide: a passage, structure or fact that came from it.

Of those 104, the audit recorded 90% with the guide link alongside it, and 9 cases with no citation at all. Credit appears most of the time, inside the answer the buyer is reading. This is not traffic, and murmur.marketing is not measuring traffic here.

Claude recommended Mindo less, Google never recommended it, and the guide produced 8 clicks in 28 days

The advance was limited to one engine. In the same cycle in which ChatGPT rose, Claude recommended Mindo less: 8 of the 123 questions it answered at both points, versus 13 before. That is 4.1 percentage points down, with 1 question entering and 6 leaving. The house rule requires publishing this decline alongside the increase, which is why it appears here rather than in a footnote.

This is all the measurement says about cause: the six Claude losses were still in the investigation queue when the report closed, without question-by-question analysis. The recorded reading is that Claude derives most of its Mindo description from the guide, to the point of transcribing whole blocks, yet still assembles its list from the institutional website. If that reading is correct, content alone will not overcome it; the answer is the longer-horizon, off-site entity work. It remains a reading, not a measured conclusion.

Google’s result was different and was not a victory either. It did not recommend Mindo once, in either June or July. In 194 of July’s 538 spontaneous captures (36%), its guide appeared as a cited source, and a source is not a recommendation. The harder cost appears in Search Console: in the 28-day window ending 10 July 2026, the entire guide property had 2,084 impressions and 8 clicks, with a 0.38% CTR. Impressions grew; traffic did not, and the operation does not imply one from the other.