Claude Brand Recommendations: Why It Skips Small Brands

Claude recommends the brands its one or two web searches surface, and its searches look for "best X tools" lists, where established vendors dominate. In 70 unbranded buyer answers, Claude named our young brand 0 times. It named the five established competitors in 21–39% of answers each. Perplexity, on the same questions, named our brand more often than any rival.

Below is the mechanism, measured: how often Claude searches, what it searches for, which pages it reads, and why it names brands without linking them. You'll finish with a diagnosis you can run on your own brand and a ranked fix list for getting into Claude's answers.

Data used: 35 unbranded buyer prompts × 2 runs on 2026-10-20 (70 Claude answers; the same prompts on ChatGPT, Gemini and Perplexity for comparison), collected by BugViso's AI Visibility Tracker for bugviso.com, a young website-audit tool, against five established, anonymised competitors (A–E). Claude ran as claude-haiku-5-5 with Anthropic's basic web search tool, capped at 3 searches per answer.

How Claude Decides What to Recommend

Claude doesn't query a fixed index the way a search engine does. Per Anthropic's web search tool documentation, "Claude determines when to search based on the prompt", the API runs the searches, and Claude writes its answer from the results. The developer can cap searches with max_uses. When Claude does search, citations are always on.

That gives four places where a brand can drop out:

1. Claude doesn't search at all and answers from what it already knows.
2. It searches, but the query is generic ("best site audit tools 2026"), so the results are lists of established vendors.
3. The results don't include your pages, or include them without your name in a recommendation-ready form.
4. It writes the answer from too few sources, and concludes that the thing you offer doesn't exist.

Our data shows all four.

The Diagnosis: Claude by the Numbers

Gemini's stored responses didn't include its search queries, so its query rows are empty. Our brand's 0 of 70 has a 95% Wilson interval of 0–5.2%. It's a real gap, not bad luck on a few answers.

1. Claude searches little

Claude ran no search at all in 17 of 70 answers, one search in 16 and two in 37. It never used the third search it was allowed. The 17 no-search answers were mostly learn-stage questions (12 of 17), and 7 of them still named tracked competitors from memory.

ChatGPT ran 3.5 searches per answer on the same prompts and Perplexity 2.8. Fewer searches mean fewer chances for a niche page to enter the answer.

2. Claude searches for lists

Of Claude's 90 search queries, 31% contained a list word: best, top, vs, comparison, alternatives or review. 24% contained a year. ChatGPT's queries had list words 2% of the time and Perplexity's 5%. Here are two of Claude's queries, with competitor names masked:

best site audit tools SEO 2026 comparison
Competitor A vs Competitor D vs Competitor C site audit limits 2026

The buyer asked "which site audit tool should I buy for a small agency that wants SEO checks plus code fixes". Claude rewrote the question as a search for roundups. Roundups rank the brands that have been around longest.

3. Claude reads lists, and lists leave new brands out

40.5% of Claude's unbranded citations (123 of 304) were list-style pages, more than any other engine. We fetched the 185 independent "best tools" pages that the engines cited for these prompts. Our brand appeared on none of them. The five competitors each appeared on 32–64%.

When a cited list named a competitor, the answer often named it too. When none of the cited lists named a competitor, the answers almost never did. Our listicles and AI recommendations study has the full breakdown.

4. Claude names without linking

Claude named competitors 98 times in unbranded answers and cited the named competitor's own website on 10 of them. ChatGPT cited the named brand's site 88% of the time, and Perplexity 81%. Claude's recommendations rest on third-party pages and on what the model already knew, not on the vendors' own pages.

That's why a strong product page doesn't move Claude by itself. On three answers, Claude even cited our blog posts as sources (a WCAG tooling guide twice and an agency tooling guide once) and still recommended other brands.

When Claude Concludes You Don't Exist

The most expensive failure is the fourth one. On 11 prompts, at least one other engine named our brand. Claude named a tracked competitor on 14 of its 22 answers to those prompts and nobody tracked on the other 8. Two answers stand out:

Both exist: a $1.49 per-report option and a custom agency plan above the 300-scan tier. When the other engines named us on these prompts, they cited our homepage or pricing page. Claude's two searches didn't reach those pages, so Claude reported the feature as missing. For a buyer, "it doesn't exist" is worse than "here are three options without you". It ends the search.

Claude does name small tools when its searches surface them. On "recommend a tool to audit a prospect's website and send a branded PDF", both runs' top pick was a small, untracked vendor that a review page had described. Claude skips the brands its searches don't find. That's a problem you can fix.

How Claude Compares on Your Own Data

You can test whether an engine recommends from search or from memory with any export of answers and citations. The signal is the share of named brands whose own site is among the citations. Low values mean the engine names brands it read about elsewhere or already knew.

#!/usr/bin/env python3
"""Does each engine recommend from search results or from memory? Uncited answers, brands named per
answer, and how often a named brand's own site is among the citations.

Usage: python3 engine_naming_profile.py answers.csv --brand Acme --domain acme.com \
         --competitor Globex --competitor-domain globex.com [...]
answers.csv columns: engine, answer, cited_urls (space-separated), branded (1/0; branded rows are skipped).
"""
import argparse, csv, re
from collections import defaultdict
from urllib.parse import urlparse

def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("csv"); ap.add_argument("--brand", required=True); ap.add_argument("--domain", required=True)
    ap.add_argument("--competitor", action="append", default=[]); ap.add_argument("--competitor-domain", action="append", default=[])
    a = ap.parse_args()
    brands = {a.brand: a.domain.lower(), **{c: d.lower() for c, d in zip(a.competitor, a.competitor_domain)}}
    rx = {b: re.compile(r"(?<![\w-])" + re.escape(b) + r"(?![\w-])", re.I) for b in brands}
    s = defaultdict(lambda: defaultdict(int))
    for r in csv.DictReader(open(a.csv, newline="", encoding="utf-8")):
        if r.get("branded") == "1":
            continue
        e = s[r["engine"]]
        hosts = {(urlparse(u).hostname or "").lower().removeprefix("www.") for u in r["cited_urls"].split()}
        e["answers"] += 1
        e["uncited"] += not hosts
        for b, d in brands.items():
            if rx[b].search(r["answer"]):
                linked = any(h == d or h.endswith("." + d) for h in hosts)
                key = "you" if b == a.brand else "rival"
                e[key + "_named"] += 1; e[key + "_linked"] += linked
    print(f"{'engine':<12}{'answers':>8}{'uncited':>9}{'brands/answer':>15}{'rivals linked':>15}{'you named':>11}{'you linked':>12}")
    for eng, e in sorted(s.items()):
        per = (e["you_named"] + e["rival_named"]) / e["answers"]
        rl = f"{100 * e['rival_linked'] / e['rival_named']:.0f}%" if e["rival_named"] else "-"
        yl = f"{e['you_linked']}/{e['you_named']}"
        print(f"{eng:<12}{e['answers']:>8}{e['uncited']:>9}{per:15.2f}{rl:>15}{e['you_named']:>11}{yl:>12}")
    print("\nLow 'rivals linked' = the engine names brands without citing their sites: it recommends from"
          "\nwhat it already knows and from third-party lists, so get onto those lists first.")

if __name__ == "__main__":
    main()

python3 engine_naming_profile.py answers.csv --brand "Acme" --domain acme.com \
  --competitor "Globex" --competitor-domain globex.com

Our data prints this:

engine       answers  uncited  brands/answer  rivals linked  you named  you linked
chatgpt           70        3           0.54            88%          6         6/6
claude            70       17           1.40            10%          0         0/0
gemini            70       11           1.43            11%          3         3/3
perplexity        70        4           0.69            81%         12       12/12

Claude and Gemini form one group: many brands per answer, almost none linked. ChatGPT and Perplexity form the other: fewer brands, nearly always linked to their own sites. A small brand can win the second group with clear pages. Winning the first group takes third-party coverage.

A Ranked Fix List for Getting Into Claude's Answers

Ranked by how many of our missed Claude answers each fix addresses:

1. Get onto the lists Claude cites. 40.5% of its citations are roundups. Find the specific pages cited for your prompts, then pitch the authors with a factual case for inclusion: what you do, for whom, at what price. This is the only fix that touches all four failure points.
2. Publish a comparison-ready page for each category question. Claude's queries include "vs", "alternatives" and the year. A page that answers "best [category] for [audience] in 2026" honestly, including competitors, matches those queries. Google's guidance on succeeding in AI search puts it as "unique, non-commodity content", and an honest comparison with real data qualifies.
3. State differentiators as plain sentences. "Unused scans roll over for 60 days" or "pay $1.49 per report, no subscription" can be quoted directly. Claude's "I couldn't find" answers came from questions where the fact existed but wasn't in a page its searches reached.
4. Earn mentions in places Claude already trusts. Documentation sites, standards bodies and well-cited guides appear in Claude's sources. A mention there reaches answers that never search for vendors.
5. Measure Claude separately. A pooled score hides a zero engine. Track Claude on its own and give each fix a few weeks before judging it.

💡 What not to do: don't rewrite your homepage for Claude. Claude linked the named brand's own site only 10% of the time, so your own pages are its least-read source on unbranded questions.

How BugViso Tracks Claude Brand Recommendations

BugViso's AI Visibility Tracker asks every buyer prompt on Claude through Anthropic's API with web search on, capped at three searches per answer, alongside ChatGPT, Gemini, Perplexity and Google AI Overviews. Each answer is stored with its search queries and citations. Brand mentions are found by name, alias and domain, then scored per engine.

For Claude specifically, you see:

• A per-engine scorecard row with mention rate, prompt coverage, citation coverage and share of voice, so a zero on Claude isn't averaged away.
• An engine segment in the leaderboard. Any segment built on fewer than 20 answers is flagged low confidence.
• Source gaps: pages cited in answers that name your competitors but not you, with the engines that cited them. That's the list to pitch from fix #1.
• The "what to do" plan, where a source-gap item comes with an AI-drafted outreach email to the page owner, and progress tracking on the next runs.

Refused or empty answers are excluded, not counted as misses, so a Claude outage doesn't look like a visibility drop. The per-engine view is shown on the AI visibility tracker page.

Traps When Reading Claude Results

Reading a no-search answer as a retrieval failure. 17 of Claude's answers ran no search. Those reflect the model's existing knowledge. A content fix won't change them quickly, but third-party coverage eventually does.

Assuming the API equals the Claude app. The app can add memory, personal settings and different search defaults. API answers approximate what a fresh session sees, so treat them as a controlled sample, not a screenshot of every user's chat.

Comparing different models. Our answers came from claude-haiku-5-5. Larger models may search more or less. Keep the model fixed within a trend and note it in the report.

Treating 0 as permanent. Our answer-consistency test found that 40% of non-empty shortlists changed when the same question was asked again 23 minutes later. One new list placement can move Claude from zero to one mention, and the visibility score's position and sentiment parts jump with it.

Blaming robots.txt first. Claude cited our blog three times, so its search could reach our site. Access matters (check it with our AI crawler robots.txt audit), but it wasn't the gap here.

Frequently Asked Questions

Why doesn't Claude recommend my brand?

Usually because its one or two searches look for "best X tools" lists and your brand isn't on them. In our data, Claude named a small brand 0 times in 70 answers, while 40.5% of its citations were list pages that left that brand out.

Does Claude search the web for every question?

No. Claude decides when to search. It ran no search on 17 of our 70 unbranded buyer questions, mostly learn-stage ones, and at most two searches on the rest, even with three allowed.

Does Claude cite company websites when it recommends a brand?

Rarely, on unbranded questions. Claude cited the named brand's own site for 10% of its competitor mentions, compared with 88% for ChatGPT. It leans on roundups and third-party pages.

How do I get Claude to mention my brand?

Get onto the specific roundups Claude cites for your buyer questions, publish honest comparison pages, and state differentiators as quotable sentences. Then measure Claude separately for several weeks.

Is Claude biased toward big brands?

Its retrieval is. Claude rewrote 31% of buyer questions into searches containing words like "best", "vs" or "alternatives", and the pages those searches return favour established vendors. Small tools that its searches did surface were recommended.

The Short Version

Claude recommends what its few, list-seeking searches find, so new brands stay invisible until the lists include them. Find the exact lists, get onto them, and watch Claude as its own channel, which is how BugViso's AI visibility tracking reports it.