Graph
Lookalike X accounts by shared audience
/data-api/v2/graph/x/{handle}/similar- Scope
- insights:read
- Freshness
- catalog read
- Group
- Graph
- Platforms
- 1 of 7
What this endpoint answers
Accounts whose audience most resembles this one, found by sampling the account's followers and reading what else they follow. Graph-neighbourhood similarity, not bio or topic matching. Ranked by lift against the panel base rate, so the answer is a real neighbourhood rather than a list of whoever is globally largest.
This is a catalog read: it is served from the CRM Solid database in milliseconds, costs one budget unit, and reports how old the reading is in meta.cache_age_s. It needs the insights:read scope (Insights): Engagement rates, top posts, viral patterns, cohort benchmarks.
Good to know
- Every count here is measured against our follow-graph panel, not against X. An edge exists because we expanded an account and read its following list, so the panel is the 477,868 accounts whose following lists we hold. @elonmusk has 39,055 indexed followers against 241M reported. Read shares and lifts, which are comparable because both sides sit on the same panel; do not read the absolute counts as real follower numbers.
- Built from at most 250 of the account's indexed followers, taken in crawl order rather than at random, then normalised against the panel base rate. sample.truncated says whether the whole audience fitted.
- Accounts with fewer than min_shared followers in the sample are not scored at all, so a very thinly indexed account can return an empty list. That is a coverage answer, not a similarity answer.
- First call for a given pair or handle reads the edge table from disk and can take several seconds; the result is then cached for 24 hours and answers in well under half a second. Sized so the worst case stays inside the read pool's 15s statement timeout.
Parameters
Every value the call accepts, with the example the spec ships so the request runs as written.
| Name | In | Required | Example | What it does |
|---|---|---|---|---|
handle | path | required | vercel | X handle without the @. 1-15 characters. |
limit | query | optional | 20 | How many lookalikes to return, 1 to 100. |
min_shared | query | optional | 6 | Minimum shared followers inside the sample before a candidate is scored, 2 to 100. Higher is stricter and returns fewer, more confident rows. |
Call it
Authenticate with a bearer token or the x-api-key header. Keys are server-to-server credentials. Never embed one in front-end code - call the API from your own backend and forward the result.
curl "https://crmsolid.com/data-api/v2/graph/x/vercel/similar" \
-H "Authorization: Bearer psk_live_..."
const res = await fetch("https://crmsolid.com/data-api/v2/graph/x/vercel/similar", {
headers: {
Authorization: `Bearer ${process.env.CRM_SOLID_DATA_API_KEY}`,
},
});
if (!res.ok) {
const { error } = await res.json();
throw new Error(`${error.code}: ${error.message} (${error.request_id})`);
}
const { data, meta } = await res.json();
import os
import requests
res = requests.get(
"https://crmsolid.com/data-api/v2/graph/x/vercel/similar",
headers={"Authorization": f"Bearer {os.environ['CRM_SOLID_DATA_API_KEY']}"},
timeout=30,
)
res.raise_for_status()
payload = res.json()
data, meta = payload["data"], payload["meta"]
Keys look like psk_live_... for production keys, psk_test_... for test keys and are minted in the panel.
What comes back
Success is { data, meta }. Failure is { error: { code, message, request_id } }. The body below is the spec's own example: the values in it are illustrative readings, not live numbers.
- Account user id
- 4686835494
- Account username
- vercel
- Account name
- Vercel
- Account followers
- 451,669
- Account is verified
- yes
- Account profile url
- https://x.com/vercel
- Coverage indexed followers estimate
- 381
- Coverage measurable
- yes
- Sample followers sampled
- 250
- Sample sample cap
- 250
- Sample truncated
- yes
- Sample min shared followers
- 4
- Sample candidates considered
- 200
- Sample candidate cap
- 200
- Panel size
- 477,868
- Panel measured at
- 2026-08-31
- Similar
- 2 items
{
"data": {
"account": {
"user_id": "4686835494",
"username": "vercel",
"name": "Vercel",
"followers": 451669,
"is_verified": true,
"profile_url": "https://x.com/vercel"
},
"coverage": {
"indexed_followers_estimate": 381,
"measurable": true
},
"sample": {
"followers_sampled": 250,
"sample_cap": 250,
"truncated": true,
"min_shared_followers": 4,
"candidates_considered": 200,
"candidate_cap": 200
},
"panel": {
"size": 477868,
"measured_at": "2026-08-31"
},
"similar": [
{
"rank": 1,
"account": {
"user_id": "1467726470533754880",
"username": "nextjs",
"name": "Next.js",
"followers": 290945,
"is_verified": true,
"profile_url": "https://x.com/nextjs",
"citation_score": 190,
"in_panel": true
},
"shared_followers": 40,
"share_of_sample_pct": 16,
"known_followers_estimate": 119,
"lift": 642.5
},
{
"rank": 2,
"account": {
"user_id": "2020202",
"username": "tailwindcss",
"name": "Tailwind CSS",
"followers": 288005,
"is_verified": true,
"profile_url": "https://x.com/tailwindcss",
"citation_score": 102,
"in_panel": false
},
"shared_followers": 12,
"share_of_sample_pct": 4.8,
"known_followers_estimate": 64,
"lift": 358.4
}
]
},
"meta": {
"request_id": "req_9f2c41a8b3d5",
"generated_at": "2026-08-23T09:14:02.317Z",
"took_ms": 42,
"page": {
"limit": 50,
"next_cursor": "eyJrIjoiMTczNDU2IiwiZCI6ImEifQ",
"count": 50
}
}
}
The meta block
request_idstringrequiredUnique id for this request. Quote it in a support ticket.
generated_atstringrequiredServer time the response was produced.
took_msintegerrequiredMilliseconds spent server-side.
pagePageoptionalsourcestringoptionalWhich backend served the payload, for endpoints with more than one.
cache_age_sintegeroptionalAge of the underlying data in seconds. 0 for live reads.
meta.page
limitintegerrequiredRows requested. Default 50, maximum 500.
next_cursorstringrequiredPass as ?cursor= to fetch the next page. null means the result set is exhausted, and it is the only reliable stop condition.
countintegerrequiredRows in this page.
totalintegeroptionalTotal matching rows. Present only when counting is cheap; never assume it is there.
List endpoints take ?limit= (max 500) and ?cursor=. Follow meta.page.next_cursor until it is null; cursors are opaque and signed.
When it fails
GET /graph/x/{handle}/similar documents 8 failure statuses. Branch on error.code, which is stable and enumerated; message is prose and may change.
- 401Unauthorized
unauthorized | invalid_keyNo key was presented, or the key is unknown, revoked or expired.
- 402PaymentRequired
payment_required | subscription_inactiveThe key is valid but the plan behind it cannot serve the call: the included requests are spent and overage is switched off, capped or unfunded (payment_required), or the billing period lapsed and was not renewed (subscription_inactive). Retrying does not help; paying does. The X-Plan-* headers on this response say how far past the line you are.
Carries X-Plan, X-Plan-Limit, X-Plan-Overage, X-Plan-Period-End, X-Plan-Remaining.
- 403Forbidden
forbidden_scope | forbidden_ipThe key is valid but not allowed to make this call: it lacks the scope, or the request came from an address outside the key's allowlist.
- 404NotFound
not_foundThe addressed resource does not exist.
- 422InvalidRequest
invalid_requestA parameter is malformed, out of range or mutually exclusive with another. `details` names the offending fields.
- 429RateLimited
rate_limited | quota_exceededEither the burst ceiling for the current minute or the daily quota is spent. Distinguish with the code: rate_limited clears within the minute, quota_exceeded does not clear until 00:00 UTC.
Carries Retry-After, X-Quota-Limit, X-Quota-Remaining, X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset, X-Request-Id.
- 500InternalError
internal_errorSomething failed on our side. Internals are never leaked; quote the request id.
- 503Unavailable
upstream_timeout | upstream_error | not_configuredThe request could not be served right now. BRANCH ON error.code, not on the status: 'upstream_timeout' means a source was too slow (this is what a catalog query hitting its 15-second statement timeout returns, so it is reachable from any endpoint that reads the corpus, not only the live-scrape ones) and the same call is worth retrying with backoff - narrowing it with a smaller limit, a filtered scope or a less popular account makes it far less likely; 'upstream_error' means a source was unreachable, so back off further; 'not_configured' means the capability has no backing service in this deployment, and retrying will never help.
Every response carries X-Request-Id and meta.request_id. Quote it in support requests.
Coverage and limits
This endpoint covers X. Rate limits come from the tier on your key.
| Tier | Requests a minute | Requests a day | Live reads a minute |
|---|---|---|---|
| free | 30 | 1,000 | 5 |
| standard | 120 | 25,000 | 20 |
| pro | 600 | 250,000 | 60 |
| unlimited | 6,000 | 10,000,000 | 600 |
This call only draws on the ordinary per-minute and per-day columns. Every response reports where you stand in X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset and the X-Plan headers.
Start calling it
A key takes a minute to mint in the panel, no card. The reference covers authentication, the envelope, scopes, rate limits and every error code in one page.
Reference path: /data-api/reference/graph-x-similar