Trustpilot Scraper — scrape Trustpilot reviews, ratings and company data

A Trustpilot scraper that pulls reviews, star ratings, company profiles and public contact details from Trustpilot without a login, an account or a cookie of yours. Give it company domains and it returns each company's profile plus its reviews. Give it a category slug and a country instead and it returns the companies ranked in that category, which is a prospect list. Output goes to an Apify dataset you can export to CSV, JSON or Excel, or read through the API. Pricing is pay per event: $0.005 per run, $0.80 per 1,000 reviews scraped, $3 per 1,000 company profiles. An empty run costs almost nothing because nothing is charged until the first result lands.

Two things make this scraper return more than a naive one. Trustpilot hides reviews written in languages other than the request language, which can turn a 200-review French company into a single visible review, and Trustpilot stops serving after 10 pages whatever the advertised page count. This Actor defaults to every language and, past 200 reviews, splits a profile into disjoint rating and language cells so each cell gets its own 10-page window. On keobiz.fr that took the harvest from 200 reviews to 749.

Table of contents

• How to scrape Trustpilot
• Which fields should I fill?
• Input reference
• Output fields
• Example output
• Scrape Trustpilot from the API
• What people use this Trustpilot scraper for
• Why this Trustpilot scraper
• Pricing
• Limits and what this Actor does not do
• Integrations
• FAQ

How to scrape Trustpilot

The Actor has two jobs and picks between them from what you fill in. Fill company domains to collect reviews, or category slugs to discover companies. Filling both is refused, because they are two different runs and guessing would waste your money.

Scrape Trustpilot reviews from the Apify Console

1. Open the Actor and go to the Input tab. www.qonto.com is already prefilled, so you can press Start immediately and look at the shape of the output before committing to anything.
2. Replace it with the companies you care about, one per line. Both a bare domain (keobiz.fr) and a full profile URL (https://www.trustpilot.com/review/www.qonto.com) are accepted, and you can mix them.
3. Set Max reviews per company. 200 is one fast pass in Trustpilot's own default order. Anything above 200 switches on the rating and language segmentation, which is slower and returns roughly three to four times more.
4. Leave Review languages on all and the proxy on residential. Those two defaults are the ones that avoid the two traps of the Trustpilot site, described further down.
5. Press Start. When the run finishes, open the Storage tab. The dataset has two views, Companies and Reviews, and an Export button for CSV, JSON, Excel and XML.

Discover Trustpilot company data by category

1. Leave Company domains empty and fill Category slugs instead, one per line: accountant, bank, insurance_agency.
2. Get a slug by opening a category on Trustpilot and taking the last part of the URL. trustpilot.com/categories/accountant gives the slug accountant. Underscores are part of the slug.
3. Set Country to an ISO code (FR, GB, US, DE) to keep only the companies of that country. Leave it empty to get every country.
4. Set Max companies per category. Trustpilot caps its own category listings, so a category frequently holds a few dozen companies whatever number you ask for. The run log prints the total Trustpilot claims for that filter so you know you got everything there was.

Getting the domain right

Trustpilot keys a profile on an exact domain string, and qonto.com and www.qonto.com can be two separate profiles with different review counts. If you are unsure, search the company on Trustpilot, open its profile, and paste the browser URL into the input. The Actor extracts the identifier from it, strips the protocol, any path and any query string, and lowercases the result.

Which fields should I fill?

If you want to…	Fill in	Example	Leave alone
Read one company's reviews	companyDomains	["keobiz.fr"]	everything else
Monitor your own Trustpilot reviews	companyDomains plus an Apify schedule	["www.qonto.com"]	everything else
Compare several competitors at once	companyDomains with many lines	["a.com","b.com","c.com"]	everything else
Go past the 200-review wall	maxReviewsPerCompany above 200	1000	everything else
Build a prospect list for an industry	categorySlugs and country	["accountant"], FR	companyDomains
Get a company's public email and phone	companyDomains (company mode only)	["wity.fr"]	—
Feed a single-language sentiment model	languages	fr	everything else
Run reliably at volume	proxyConfiguration on residential	default	everything else

One of the two entry fields is mandatory. Everything else already has a working default, and an empty input fails immediately with a message telling you which field to fill rather than a stack trace.

Input reference

Field	Type	Default	What it does
companyDomains	array of strings	empty	Domains or Trustpilot profile URLs to scrape reviews and company data from. One per line.
categorySlugs	array of strings	empty	Trustpilot category slugs to discover companies from. Mutually exclusive with companyDomains.
country	string	empty	ISO country code filtering category discovery. Ignored in company mode.
mode	string enum	auto	auto, company or category. Auto reads the mode off the filled fields; set it by hand only when a saved task keeps both fields filled.
maxReviewsPerCompany	integer	200	Upper bound on reviews per company. Above 200 the rating and language segmentation kicks in.
maxCompaniesPerCategory	integer	100	Upper bound on companies per category, paged 20 at a time.
languages	string	all	all for every review language, or a single ISO code such as fr or en.
proxyConfiguration	object	Apify residential	Proxy settings. Residential is the default and the setting that matters most here.

Example input, company reviews:

{
  "companyDomains": ["www.qonto.com", "keobiz.fr"],
  "maxReviewsPerCompany": 1000,
  "languages": "all",
  "proxyConfiguration": {
    "useApifyProxy": true,
    "apifyProxyGroups": ["RESIDENTIAL"]
  }
}

Example input, category discovery:

{
  "categorySlugs": ["accountant"],
  "country": "FR",
  "maxCompaniesPerCategory": 50
}

Output fields

Every record carries a type field, either company or review, and the dataset ships two views so the Console shows them as two separate tables.

Company record, company mode

This is the full Trustpilot company profile, emitted once per domain before its reviews.

Field	Type	Description	Example
type	string	Always company for this record type	"company"
id	string	Trustpilot's internal business unit id	"5f8c1b0d9f4a2c0001b7e3a1"
domain	string	The domain Trustpilot keys the profile on	"keobiz.fr"
name	string	Company display name	"Keobiz"
trustScore	number	TrustScore out of 5, one decimal	4.6
stars	number	Star bucket Trustpilot displays	4.5
numberOfReviews	integer	Total reviews Trustpilot advertises	4512
reviewsLast12Months	integer	Reviews published in the last 12 months	1189
isClaimed	boolean	Whether the company claimed its profile	true
isClosed	boolean	Whether Trustpilot marked the profile closed	false
countryCode	string	Country of the business unit	"FR"
categories	array of objects	Categories with id, name, isPrimary	[{"id":"accountant","name":"Accountant","isPrimary":true}]
websiteUrl	string	Company website as declared on Trustpilot	"https://www.keobiz.fr"
email	string or null	Public contact email from the profile	"contact@keobiz.fr"
phone	string or null	Public contact phone	"+33 1 76 41 04 88"
address	string or null	Street address	"12 rue de la Chaussée d'Antin"
city	string or null	City	"Paris"
zipCode	string or null	Postal code	"75009"
country	string or null	Country from the contact block	"France"

The contact fields are null when the company never published them. They are the reason this Actor doubles as a lead source: a Trustpilot profile with an email and a phone number is a qualified company record, not just a rating.

Company record, category discovery mode

Discovery rows come from the category listing, which carries less than a profile page. There is no contact block here.

Field	Type	Description	Example
type	string	Always company	"company"
domain	string	Company domain, usable as input for a review run	"wity.fr"
name	string	Company display name	"Wity"
trustScore	number	TrustScore out of 5	4.8
stars	number	Star bucket	5
numberOfReviews	integer	Total reviews Trustpilot advertises	612
city	string or null	City from the listing	"Paris"
country	string or null	Country from the listing	"FR"
categoryId	string	The slug you asked for	"accountant"

The usual pipeline is two runs: discover the companies of a category, then feed the domain column back in as companyDomains to get contact details and reviews.

Review record

Field	Type	Description	Example
type	string	Always review	"review"
id	string	Trustpilot review id, stable across runs	"66b3f0c21e4ab90c2d5f8e71"
companyDomain	string	Domain of the company the review belongs to	"keobiz.fr"
rating	integer	Star rating from 1 to 5	5
title	string	Review headline	"Fast and clear answers"
text	string	Full review body	"Onboarding took two days and…"
language	string	ISO code of the language the review is written in	"fr"
experiencedDate	string, ISO 8601	Date of the experience the reviewer describes	"2026-07-14T00:00:00.000Z"
publishedDate	string, ISO 8601	When the review was published	"2026-07-16T09:12:44.000Z"
updatedDate	string or null	When the reviewer last edited it	null
likes	integer	Useful votes the review received	3
isVerified	boolean	Trustpilot's verification label	true
consumerName	string	Reviewer display name	"Camille D."
consumerCountry	string	Reviewer country code	"FR"
consumerReviewCount	integer	How many reviews that account has written	7
replyMessage	string or null	The company's public reply	"Thanks Camille, glad it went well."
replyDate	string or null	When the company replied	"2026-07-17T08:03:10.000Z"

replyMessage is what makes response-rate reporting possible: count the reviews under 3 stars where it is null and you have the backlog of unanswered complaints.

Example output

One company record and one review record, as they land in the dataset:

[
  {
    "type": "company",
    "id": "5f8c1b0d9f4a2c0001b7e3a1",
    "domain": "keobiz.fr",
    "name": "Keobiz",
    "trustScore": 4.6,
    "stars": 4.5,
    "numberOfReviews": 4512,
    "reviewsLast12Months": 1189,
    "isClaimed": true,
    "isClosed": false,
    "countryCode": "FR",
    "categories": [
      { "id": "accountant", "name": "Accountant", "isPrimary": true },
      { "id": "business_management_consultant", "name": "Business Management Consultant", "isPrimary": false }
    ],
    "websiteUrl": "https://www.keobiz.fr",
    "email": "contact@keobiz.fr",
    "phone": "+33 1 76 41 04 88",
    "address": "12 rue de la Chaussée d'Antin",
    "city": "Paris",
    "zipCode": "75009",
    "country": "France"
  },
  {
    "type": "review",
    "id": "66b3f0c21e4ab90c2d5f8e71",
    "companyDomain": "keobiz.fr",
    "rating": 5,
    "title": "Fast and clear answers",
    "text": "Onboarding took two days and every question I sent got an answer the same afternoon.",
    "language": "fr",
    "experiencedDate": "2026-07-14T00:00:00.000Z",
    "publishedDate": "2026-07-16T09:12:44.000Z",
    "updatedDate": null,
    "likes": 3,
    "isVerified": true,
    "consumerName": "Camille D.",
    "consumerCountry": "FR",
    "consumerReviewCount": 7,
    "replyMessage": "Thanks Camille, glad it went well.",
    "replyDate": "2026-07-17T08:03:10.000Z"
  }
]

And a discovery row from category mode:

{
  "type": "company",
  "domain": "wity.fr",
  "name": "Wity",
  "trustScore": 4.8,
  "stars": 5,
  "numberOfReviews": 612,
  "city": "Paris",
  "country": "FR",
  "categoryId": "accountant"
}

Scrape Trustpilot from the API

The Actor runs like any Apify Actor, so it is reachable from the Apify API, from the official clients, or from plain HTTP. Replace <YOUR_API_TOKEN> with the token on your Apify account settings page.

Python

from apify_client import ApifyClient

client = ApifyClient("<YOUR_API_TOKEN>")

run_input = {
    "companyDomains": ["www.qonto.com", "keobiz.fr"],
    "maxReviewsPerCompany": 1000,
    "languages": "all",
    "proxyConfiguration": {
        "useApifyProxy": True,
        "apifyProxyGroups": ["RESIDENTIAL"],
    },
}

run = client.actor("alexmorain/trustpilot-scraper").call(run_input=run_input)

reviews, companies = [], []
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    (companies if item["type"] == "company" else reviews).append(item)

print(f"{len(companies)} companies, {len(reviews)} reviews")
for r in reviews[:5]:
    print(r["rating"], r["publishedDate"], r["title"])

JavaScript

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: '<YOUR_API_TOKEN>' });

const input = {
    categorySlugs: ['accountant'],
    country: 'FR',
    maxCompaniesPerCategory: 50,
};

const run = await client.actor('alexmorain/trustpilot-scraper').call(input);
const { items } = await client.dataset(run.defaultDatasetId).listItems();

const companies = items.filter((i) => i.type === 'company');
console.log(`${companies.length} companies discovered`);
for (const c of companies) {
    console.log(c.name, c.domain, c.trustScore, c.numberOfReviews);
}

cURL

Start a run and wait for it to finish, then read the dataset:

curl -X POST "https://api.apify.com/v2/acts/alexmorain~trustpilot-scraper/run-sync-get-dataset-items?token=<YOUR_API_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{
    "companyDomains": ["keobiz.fr"],
    "maxReviewsPerCompany": 200,
    "languages": "all"
  }'

Same call, but asking for a CSV export of the reviews view only:

curl -X POST "https://api.apify.com/v2/acts/alexmorain~trustpilot-scraper/run-sync-get-dataset-items?token=<YOUR_API_TOKEN>&format=csv&view=reviews" \
  -H "Content-Type: application/json" \
  -d '{"companyDomains": ["keobiz.fr"], "maxReviewsPerCompany": 500}' \
  -o trustpilot-reviews.csv

That second command is the shortest path to export Trustpilot reviews to CSV: one request, one file, no post-processing.

What people use this Trustpilot scraper for

Trustpilot review monitoring on your own company

Put your own domain in companyDomains, cap reviews at 200, and schedule the run daily on Apify. Each run returns reviews in Trustpilot's own profile order, with rating, publishedDate and replyMessage, so a small script can compare against yesterday's dataset and alert on anything new below three stars, or on anything older than 48 hours that still has a null replyMessage. Trustpilot's own notification emails tell you a review landed; a dataset tells you how the backlog of unanswered ones is trending, which is the number that actually moves your score.

Trustpilot competitor analysis

Run several domains in one go and you get a comparable table: trustScore, numberOfReviews, reviewsLast12Months and the full review text for each. The 12-month count next to the total is the useful pair, because it separates a competitor with a good historical score from one that is currently collecting reviews. Reading the one and two star reviews of a competitor is the cheapest customer research there is: people describe, in their own words and for free, exactly which promise the competitor failed to keep. Those sentences go straight into positioning and into sales objection handling.

Trustpilot lead generation from company profiles

Category discovery returns the companies ranked in an industry and a country, and a second pass over those domains returns each profile's public email, phone and postal address. That is a prospect list built on a signal you cannot buy from a database vendor: these companies care enough about their reputation to have a claimed Trustpilot profile, and their review volume tells you how large and how active they are. Filtering the output on trustScore below 4 gives you the subset with a reputation problem, which is a conversation opener if you sell anything touching customer experience.

Feeding a sentiment or topic model

Reviews come out with the body text, the star rating, the language code and both dates, which is the shape a supervised model wants: the rating is the label, the text is the input, the date lets you window the training set. Because the language filter defaults to all, a multilingual company's corpus stays multilingual instead of collapsing to whatever language Trustpilot decided to serve. If your downstream model is monolingual, set languages to that ISO code and the Actor filters at the source rather than making you throw rows away afterwards.

Due diligence on a supplier or an acquisition target

Before signing with a vendor, pull their full review history and look at the distribution rather than the headline score. A 4.5 built on 3,000 reviews with a steady flow reads differently from a 4.5 built on 60 reviews that all landed in the same fortnight, and the record gives you both numbers. isClaimed, isClosed and the reply rate say how the company handles public criticism, which is a decent proxy for how it will handle your complaint once you are a customer.

Market and category research

Discovery mode over several category slugs and several countries produces a map of who exists in a market, with a rating and a review volume attached to each. It answers the boring questions that usually take a week of manual work: how many players are there, how concentrated is the review volume, which countries have an active local market and which do not.

Why this Trustpilot scraper

The language filter that silently empties a profile. Trustpilot returns only reviews written in the language of the request. A French company with 200 reviews can come back with a single one, and nothing in the output tells you the other 199 were dropped. It looks like a company with no reviews, not like a bug, which is why it survives in scrapers for a long time. This Actor requests every language by default and only narrows it when you explicitly ask.

The 10-page wall. Trustpilot advertises thousands of pages on large profiles and serves at most ten of them, twenty reviews each. Any scraper that just increments ?page= stops at 200 reviews per company. This one detects that ceiling and splits the profile into disjoint cells by star rating and by language, using the exact per-rating counts Trustpilot publishes in its own filter statistics, then walks each cell's own 10-page window and de-duplicates by review id. On keobiz.fr that returned 749 reviews where a plain paginated scrape returned 200.

No login, no account, no cookies of yours. Everything the Actor reads is server-rendered JSON that Trustpilot embeds in its own public pages. You never hand over credentials, and there is no account of yours to get suspended. A headless browser is used for one thing only, solving the bot-check challenge to obtain an access token, after which all the actual scraping is plain HTTP. That token is cached in a named key-value store between runs, so most runs never start a browser at all. That is why the measured platform cost dropped from $2.43 to $0.32 per 1,000 results, and it is why the per-review price can sit where it does.

Public contact details, not just ratings. The company record carries the email, phone and postal address block that Trustpilot publishes on the profile. Most Trustpilot scrapers return reviews and stop. Those three fields are what turns a review dataset into a prospect list.

Discovery by category and country. You do not need to know the domains in advance. Give a category slug and a country code and the Actor returns the ranked companies, which you then feed back in as input. Scrapers that only accept URLs assume you already have the list, which is usually the hard part.

Failure that stays local. A company that stays blocked is logged with an explanation and skipped; the run keeps the reviews it already collected and moves to the next domain. The run only ends in failure when every single target failed, and it says so in the status message instead of exiting green on an empty dataset.

Pricing

Pay per event. You are charged for what the Actor delivers, and nothing is charged until the first result is pushed, so a run that gets blocked from end to end costs you nothing.

Event	Price	When it fires
run-started	$0.005 per run	Once, on the first delivered result. Covers the fixed per-run cost.
apify-actor-start	$0.00005 per GB of memory	Once at container start, per gigabyte allocated.
review-scraped	$0.0008 per review, so $0.80 per 1,000	Each review pushed to the dataset.
company-scraped	$0.003 per company, so $3 per 1,000	Each company record, profile or discovery row.

Worked examples:

One company, 200 reviews. $0.005 for the run, $0.003 for the profile, $0.16 for the reviews. Total $0.168.

One company, deep pass with maxReviewsPerCompany at 1000, returning 750 reviews. $0.005 plus $0.003 plus $0.60. Total $0.608.

Twenty competitors, 200 reviews each, 4,000 reviews in one run. $0.005 for the run, $0.06 for the twenty profiles, $3.20 for the reviews. Total $3.265.

5,000 reviews collected across a batch of companies. $4.00 in review events, plus $0.003 per company profile and the single $0.005 run fee. A 25-company batch lands at roughly $4.08.

Category discovery, 100 companies. $0.005 for the run and $0.30 for the hundred rows. Total $0.305. Contact details are not in these rows; a second run over the discovered domains adds $0.003 per company for the full profile.

Apify's own compute and residential proxy usage is absorbed in these prices, so the numbers above are what you pay. If you set a maximum charge on the run, the Actor stops as soon as it hits the limit and says so in the status message, rather than continuing to work and getting cut off.

Limits and what this Actor does not do

Trustpilot never serves more than 200 reviews for a single request. The rating and language segmentation pushes that wall back, typically to three or four times more, but it does not remove it. A profile advertising 58,000 reviews will not come back with 58,000 reviews, from this Actor or any other, because Trustpilot does not serve them. The run log prints what was collected against the advertised total so you always know where you stand.

Category listings are capped by Trustpilot itself. accountant filtered on France lists a couple of dozen companies, not every accountant in the country. Raising maxCompaniesPerCategory does not conjure companies Trustpilot does not rank.

Category discovery rows do not carry email, phone or address. Those come from the profile page, so getting contact details for discovered companies means a second run in company mode using the domain column.

There is no incremental mode. Every run scrapes from scratch and de-duplication happens within a run, not across runs. Comparing against a previous dataset is on you; review id is stable and is the field to join on.

There is no keyword search over reviews, no date range filter, and no sort order option. The Actor takes what Trustpilot's own filters give it: language and star rating. Filter on publishedDate after export.

Reviewer profile URLs, review images, and Trustpilot's invitation metadata are not extracted. What you get is the field list above, nothing more.

Datacenter proxies or no proxy will get large runs blocked. Trustpilot rate-limits by IP, and a single IP holds for roughly 40 to 80 requests before it starts returning 403 whatever token it carries. Residential is the default for that reason, and changing it is the one input change that can break a run.

Scrape public data only, and use it in line with Trustpilot's terms and the law that applies to you. Personal data in reviews is personal data; if you store it, GDPR and its equivalents apply to you the same way they would if you had typed it in yourself.

Integrations

Runs and datasets are Apify platform objects, so everything the platform offers applies here. Schedule a run daily, hourly or on any cron expression from the Schedules tab. Attach a webhook that fires on ACTOR.RUN.SUCCEEDED and posts the run id and dataset id to your endpoint, which is the clean way to build Trustpilot review monitoring that pushes to Slack or to your own database instead of being polled.

Datasets export to JSON, JSON Lines, CSV, Excel, XML, HTML and RSS, either from the Console Export button or by adding &format=csv to the dataset items API call. The view=companies and view=reviews query parameters split the two record types without any client-side filtering, so a bulk Trustpilot export straight into a spreadsheet is a single URL.

The Actor is also reachable from Apify's MCP server, which lets an agent call it as a tool and read the results back. Combined with the official apify-client packages for Python and JavaScript shown above, and Apify's connectors for the usual automation platforms, wiring Trustpilot company data into an existing pipeline is a configuration job rather than a development one.

FAQ

Can I scrape Trustpilot without an account? Yes. The Actor never logs in and never uses cookies belonging to you. It reads the JSON that Trustpilot server-renders into its own public pages. You need an Apify account to run the Actor, nothing on the Trustpilot side.

Is scraping Trustpilot legal? Collecting publicly available data is generally lawful in the EU and the US, but that is not blanket permission. Trustpilot's terms of service, your local law and data protection rules such as the GDPR still apply, especially since reviews contain personal data like reviewer names and countries. Use the output for analysis, keep it secure, and take legal advice if you plan to republish it or use it for direct marketing. This is not legal advice.

How many reviews can I export per company? Up to 200 in a single fast pass. Above that, the segmentation typically brings back three to four times more, measured at 749 on a profile where plain pagination gave 200. Very large profiles cannot be exported in full, because Trustpilot itself stops serving beyond a certain depth.

Why do I only get one review for a French or German company? That is Trustpilot's language filter, and it is exactly the trap this Actor works around. Check that languages is set to all. If you set it to a specific language code, you get only reviews written in that language, which on a non-English profile can be almost nothing.

How do I export Trustpilot reviews to CSV? Run the Actor, open the Storage tab, choose the Reviews view and press Export, or call the API with format=csv&view=reviews as shown in the cURL example above. Excel, JSON and XML work the same way.

Can I scrape all companies in a Trustpilot category? You get the companies Trustpilot ranks for that category and country filter, which is capped by Trustpilot, not by the Actor. That is often a few dozen for a narrow national category. The run log prints the total Trustpilot reports so you can tell a cap from a bug.

Does it return the company's email and phone number? Yes, in company mode, when the company published them on its Trustpilot profile. email, phone, address, city and zipCode come from the profile's public contact block and are null when the company left them empty. Category discovery rows do not include them.

Do I need proxies to scrape Trustpilot? Yes, for anything beyond a few dozen requests. Trustpilot rate-limits by IP and a single address gets blocked after roughly 40 to 80 requests. Apify residential proxies are the default input and the proxy cost is included in the price.

Can I monitor new Trustpilot reviews automatically? Yes, with an Apify schedule plus a webhook. A daily run capped at 200 reviews covers the recent window of all but the highest-volume profiles. Compare review id values against your previous run to isolate what is new.

How much does it cost to scrape 1,000 Trustpilot reviews? $0.80 in review events, plus $0.005 for the run and $0.003 per company profile. A single-company run of 1,000 reviews comes to about $0.81.

What happens if Trustpilot blocks one of my companies? That company is logged with the reason and skipped. Reviews already collected for it are kept, and the run moves on to the next domain. The run only fails outright if every target failed.

Can I scrape Trustpilot reviews by keyword or date range? Not from the input. The Actor exposes the filters Trustpilot itself supports, which are language and star rating. Reviews carry publishedDate and text, so filter after export.

Which Trustpilot domain does it use? www.trustpilot.com. Country-specific domains like fr.trustpilot.com are accepted in the input as profile URLs, and the Actor extracts the company identifier from them.

Can I use this Actor from an AI agent? Yes, through Apify's MCP server or the standard API. The input schema is small and every field has a description, which is usually enough for an agent to call it correctly on the first try.