Skip to content

Build a brand co-mention matrix from research titles

Calculate which brand names occur together in captured research titles, with a reproducible matrix, pair counts and explicit limits on interpretation.

Guides

Which named brands appear together in the same returned titles?

A mention-share calculation counts how often each brand appears, but does not show whether the names occur in the same records. A co-mention matrix answers that narrower question. It can help you assemble a reading queue for product comparisons, provided you keep the original links and inspect the content before treating a named pair as an actual comparison. This guide computes the matrix from genuine returned titles; it does not measure shared audiences, purchases, creator relationships or all online discussion.

Title co-mention analysis brief
Using only the captured public research titles, build a symmetric co-mention matrix for Ninja, Lidl and Cuisinart. Count each exact original URL once; stop if one URL has conflicting titles. Match normalized whole-word brand names, report reviewed-record and unmatched-title counts, and give each unordered pair's count and percentage of reviewed URLs. Preserve original evidence. Do not infer video contents, shared audiences, market share or sentiment.
01 / Guides

Define a fixed title dictionary

Choose the literal brand names before calculating the matrix. The captured example uses Ninja, Lidl and Cuisinart in a frozen research response about ice-cream machines. The implementation lowercases names and titles, replaces punctuation with spaces, and matches complete normalized word sequences. It does not infer abbreviations, translate names or resolve ambiguous entities. If a useful title names a product but omits its manufacturer, record that limitation rather than silently editing the title or expanding the dictionary after seeing the result.

02 / Guides

Fix the counting unit and reconcile repeats

Use one record per exact original URL in the public response. Repeated identical URLs with identical titles count once; conflicting titles for one URL stop this calculation instead of silently selecting a convenient version. A title that is just its own URL, with or without the scheme, is retained as unusable and excluded from the reviewed-title denominator; it is not a confirmed zero-brand title. Different URL spellings can still refer to the same video, so this rule is exact-URL deduplication, not verified video-identity resolution. Retain the raw response and reconcile known aliases separately if your claim needs a count of distinct videos or creators.

03 / Guides

Build one presence vector per record

For each retained title, create a true-or-false value for every dictionary brand. A repeated name within one title remains one presence, not several mentions. A title can contain no dictionary names, one name or multiple names. Keep titles with no named brand in the reviewed-record denominator and report their count. Removing them without saying so would change what the pair percentages describe.

04 / Guides

Fill the diagonal and paired cells

Each diagonal cell counts titles naming that one brand. Each off-diagonal cell counts titles naming both its row and column brands. The matrix is symmetric because an unordered pair has the same count in either direction. A title naming three brands contributes to three different pairs, so adding all pair counts does not produce a count of unique comparison videos. The program also emits each unordered pair once to avoid double-counting its two mirrored cells.

05 / Guides

Use the reviewed-title denominator

For each pair, divide its count by the number of unique exact URLs reviewed, not by the sum of brand mentions used in a share-of-voice calculation. The program rounds the displayed percentage to two decimal places and leaves a zero-denominator result undefined. A zero pair count means no retained title matched both literal names under this rule; it does not establish that the brands were never compared in the actual footage, other records or the wider market.

06 / Guides

Replay the calculation on the saved response

Run the local code below without a network connection to reproduce the stored output from the embedded public records. The request and API-response blocks remain separate from the derived matrix. To study a fresh response, save its actual request, capture time and returned rows as a new observation before calculating again. A changed matrix can reflect a different retrieval sample, changed titles or changed dictionary terms, not necessarily a change in public attention.

07 / Guides

Turn co-mentions into a review queue, not a market conclusion

Open the original sources for a pair you want to investigate and establish what the clips actually compare. A title can be promotional, rhetorical, inaccurate or unrelated to the video's main argument. Keep unavailable sources unresolved. The matrix alone cannot establish comparative performance, sponsorship, sentiment, audience overlap or demand. Compare periods only when the query, dictionary, URL identity rule and source conditions remain compatible, and disclose retrieval gaps beside the table.

Captured usage and output

Captured 2026-11-18T07:26:44.240Z through the signed-in App HTTP API. HTTP 200; result status: partial. This is a recorded read-only retrieval, not a live widget or proof that this entire workflow, tracking or a particular MCP client was verified. Results can change.

POST https://app.montages.co/api/tools/research. Use the current authentication requirements when reproducing the request.

Exact research request body
{
  "query": "Ninja Creami Lidl Creami Cuisinart FastFreeze comparison review YouTube",
  "media_type": "video",
  "limit": 5
}
Runnable request · signed-in app browser console
if (location.origin !== "https://app.montages.co") throw new Error("Run this in your signed-in Montage app tab");
const response = await fetch("https://app.montages.co/api/tools/research", {
  method: "POST",
  credentials: "same-origin",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({
  "query": "Ninja Creami Lidl Creami Cuisinart FastFreeze comparison review YouTube",
  "media_type": "video",
  "limit": 5
})
});
if (!response.ok) throw new Error(`HTTP ${response.status}`);
console.log(await response.json());
Captured API response · public fields only
{
  "results": {
    "videos": [
      {
        "title": "We Tasted Ice Cream from 5 Ice Cream Makers (Cuisinart, Ninja CREAMi, Whynter) | Gear Heads",
        "url": "https://youtube.com/watch?v=U6tsJ79ghg4",
        "source": "youtube.com",
        "published_at": "2026-07-06T00:37:55.000Z"
      },
      {
        "title": "ALDI Eismaschine für 99,99€ – Ninja Creami jetzt überflüssig?",
        "url": "https://www.youtube.com/watch?v=H1alpCVurOI",
        "source": "www.youtube.com"
      },
      {
        "title": "instagram.com/reel/Db5n1snE_8e",
        "url": "https://instagram.com/reel/Db5n1snE_8e",
        "source": "instagram.com",
        "published_at": "2026-08-11T12:47:10.375Z"
      },
      {
        "title": "Is the Ninja Creami worth the hype? @alainachou put it head-to-head with the new smaller and… | TikTok",
        "url": "https://tiktok.com/@bonappetitmag/video/7644308295775210765",
        "source": "tiktok.com",
        "published_at": "2026-05-26T21:00:20.000Z"
      },
      {
        "title": "I Tested and Ranked Every Protein Ice Cream Machine (Ninja Creami's vs Competitors)",
        "url": "https://youtube.com/watch?v=UPDna1Cf6Ok",
        "source": "youtube.com"
      }
    ]
  },
  "meta": {
    "status": "partial"
  }
}

Reproduce the local calculation. This code ran on the captured public metadata below. Its output is a local transformation, not another API response or proof of what a video says. Copy and run it with Node.js; it needs no credentials or paid collection.

Runnable Node.js calculation on captured records
const response = {
  "results": {
    "videos": [
      {
        "title": "We Tasted Ice Cream from 5 Ice Cream Makers (Cuisinart, Ninja CREAMi, Whynter) | Gear Heads",
        "url": "https://youtube.com/watch?v=U6tsJ79ghg4",
        "source": "youtube.com",
        "published_at": "2026-07-06T00:37:55.000Z"
      },
      {
        "title": "ALDI Eismaschine für 99,99€ – Ninja Creami jetzt überflüssig?",
        "url": "https://www.youtube.com/watch?v=H1alpCVurOI",
        "source": "www.youtube.com"
      },
      {
        "title": "instagram.com/reel/Db5n1snE_8e",
        "url": "https://instagram.com/reel/Db5n1snE_8e",
        "source": "instagram.com",
        "published_at": "2026-08-11T12:47:10.375Z"
      },
      {
        "title": "Is the Ninja Creami worth the hype? @alainachou put it head-to-head with the new smaller and… | TikTok",
        "url": "https://tiktok.com/@bonappetitmag/video/7644308295775210765",
        "source": "tiktok.com",
        "published_at": "2026-05-26T21:00:20.000Z"
      },
      {
        "title": "I Tested and Ranked Every Protein Ice Cream Machine (Ninja Creami's vs Competitors)",
        "url": "https://youtube.com/watch?v=UPDna1Cf6Ok",
        "source": "youtube.com"
      }
    ]
  },
  "meta": {
    "status": "partial"
  }
};
const output = (function countTitleCoMentions(response, brands) {
  const byUrl = new Map();
  for (const row of Object.values(response.results).flat()) {
    if (byUrl.has(row.url) && byUrl.get(row.url).title !== row.title) throw new Error('Conflicting titles for one original URL');
    byUrl.set(row.url, row);
  }
  const normalize = value => value.toLowerCase().replace(/[^a-z0-9]+/g, ' ').trim();
  const allRows = [...byUrl.values()];
  const isUrlTitle = row => row.title.trim().replace(/^https?:\/\//i, '').replace(/\/$/, '') === row.url.replace(/^https?:\/\//i, '').replace(/\/$/, '');
  const unusableTitles = allRows.filter(isUrlTitle).map(row => ({ url: row.url, title: row.title, reason: 'URL used as title; brand presence is unresolved.' }));
  const rows = allRows.filter(row => !isUrlTitle(row));
  const present = rows.map(row => brands.map(brand => (' ' + normalize(row.title) + ' ').includes(' ' + normalize(brand) + ' ')));
  const matrix = brands.map((brand, left) => brands.map((other, right) => present.filter(flags => flags[left] && flags[right]).length));
  const pairs = brands.flatMap((brand, left) => brands.slice(left + 1).map((other, offset) => {
    const titlesWithBoth = matrix[left][left + offset + 1];
    return { brands: [brand, other], titlesWithBoth, percentOfReviewedTitles: rows.length === 0 ? null : Math.round(titlesWithBoth * 10000 / rows.length) / 100 };
  }));
  return { unit: 'One record per exact original URL with a usable title, using normalized literal title terms only.', uniqueUrlsReturned: allRows.length, recordsReviewed: rows.length, unusableTitles, brands, matrix, pairs,
    titlesWithNoNamedBrand: present.filter(flags => !flags.some(Boolean)).length,
    limitation: 'Co-occurring title names do not establish a substantive comparison, shared viewers, purchases, sentiment or population-level attention.' };
})(response, ["Ninja","Lidl","Cuisinart"]);
console.log(JSON.stringify(output, null, 2));
Actual local calculation output
{
  "unit": "One record per exact original URL with a usable title, using normalized literal title terms only.",
  "uniqueUrlsReturned": 5,
  "recordsReviewed": 4,
  "unusableTitles": [
    {
      "url": "https://instagram.com/reel/Db5n1snE_8e",
      "title": "instagram.com/reel/Db5n1snE_8e",
      "reason": "URL used as title; brand presence is unresolved."
    }
  ],
  "brands": [
    "Ninja",
    "Lidl",
    "Cuisinart"
  ],
  "matrix": [
    [
      4,
      0,
      1
    ],
    [
      0,
      0,
      0
    ],
    [
      1,
      0,
      1
    ]
  ],
  "pairs": [
    {
      "brands": [
        "Ninja",
        "Lidl"
      ],
      "titlesWithBoth": 0,
      "percentOfReviewedTitles": 0
    },
    {
      "brands": [
        "Ninja",
        "Cuisinart"
      ],
      "titlesWithBoth": 1,
      "percentOfReviewedTitles": 25
    },
    {
      "brands": [
        "Lidl",
        "Cuisinart"
      ],
      "titlesWithBoth": 0,
      "percentOfReviewedTitles": 0
    }
  ],
  "titlesWithNoNamedBrand": 0,
  "limitation": "Co-occurring title names do not establish a substantive comparison, shared viewers, purchases, sentiment or population-level attention."
}

Public titles, original URLs, source hostnames, available publication dates and result status only. Private fields, credentials and continuation cursors are omitted. Source content is not independently verified. Capture integrity checksum: 8c62bb6cc31f282e59d7b4c08b10332b82b7deab928f76cb94400653cb7036c4; the checksum checks file integrity, not source truth.

Keep the boundaries clear.

  • Every matrix cell counts literal title matches, not verified video contents.
  • The diagonal, unordered pair counts and mention-share percentages use different counting rules; do not sum them as independent populations.
  • The captured research status and source limits still apply to the derived matrix.
  • This actual partial response mixes YouTube, TikTok and Instagram. The Instagram row contains only a URL as its title and is excluded as unusable, leaving four usable titles. The Ninja–Cuisinart pair appears in one of those four titles (25 percent); no usable title names Lidl. Neither result establishes anything about the wider market.
  • TikTok source availability and all video contents remain unverified; title counts describe only the returned metadata.

Start with a read-only check: ask your agent to report actual source coverage before collecting records or exporting an audience.

Primary references

Documentation reviewed 2026-11-18. Product capabilities, client availability and plans can change. Third-party marks identify their products, not an affiliation.

Know what social media shows about you.
Start with a question.

Get started
montage