{"id":380,"date":"2026-08-16T14:33:00","date_gmt":"2026-08-16T14:33:00","guid":{"rendered":"https:\/\/texttoolz.com\/blog\/?p=380"},"modified":"2026-08-16T14:33:01","modified_gmt":"2026-08-16T14:33:01","slug":"data-annotation-companies","status":"publish","type":"post","link":"https:\/\/texttoolz.com\/blog\/data-annotation-companies\/","title":{"rendered":"What Every Company Should Know Before Buying AI Training Data"},"content":{"rendered":"

Search for data annotation companies and Google returns two different markets at once: the companies selling annotation, and the people doing it. The hourly rate is published only on the second group’s pages, and it is the closest thing to a price anybody publishes in this category.<\/strong><\/p>\n

Every buyer-facing list ranking for this question is published by a company that sells in the category. Lightly ranks Lightly first. Voxel51 and Encord both sell annotation platforms. RWS sells TrainAI. Label Your Data is a vendor’s own home page. The one directory that ranks discloses that it carries sponsored placements.<\/p>\n

TextToolz sells no annotation, no labelling tooling and no dataset, and takes no referral fee from anything named here. That is the only reason this page can put the buyer-side and worker-side numbers on the same screen.<\/p>\n

What data annotation actually is<\/h2>\n

Data annotation is the work of labelling raw material so a model can learn from it: drawing boxes around objects in images, transcribing and tagging audio, marking entities in text, segmenting LiDAR point clouds, or ranking two model outputs against each other. The label is the signal; the raw file on its own teaches nothing.<\/p>\n

On the question of whether “data labeling” and “data annotation” mean different things, the honest answer is visible in the sources themselves: this entire result set uses them interchangeably, sometimes in the same sentence. Treat any page that insists on a sharp distinction as making a positioning argument rather than a technical one.<\/p>\n

The split that does matter runs through everything below. You are either buying a managed service, where somebody else’s workforce does the work to your spec, or buying tooling your own team operates. Several companies on this search sell both, and a buyer who does not know which one they are in a conversation about will get quoted for the wrong thing.<\/p>\n

The only prices anyone publishes<\/h2>\n

Two rate cards exist across this entire search. They sit on opposite sides of the transaction.<\/p>\n\n\n\n\n\n\n\n\n
Published rates for annotation work, both sides of the market, read 15 August 2026.<\/caption>\n
Source<\/th>\nPublished rate<\/th>\nSide of the transaction<\/th>\n<\/tr>\n<\/thead>\n
DataAnnotation.tech<\/td>\n$25 to $30+ per hour, generalist<\/td>\nPaid to the worker<\/td>\n<\/tr>\n
DataAnnotation.tech<\/td>\n$50 to $100+ per hour<\/strong> for coding, accounting, finance, mathematics, physics, chemistry, biology, law and medicine<\/td>\nPaid to the worker<\/td>\n<\/tr>\n
DataAnnotation.tech<\/td>\n$20 to $50+ per hour, bilingual<\/td>\nPaid to the worker<\/td>\n<\/tr>\n
GoodFirms directory<\/td>\n$25 to $49 per hour (its top-ranked firm, ELEKS)<\/td>\nCharged to the buyer<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

The specialist domains reach $50 to $100 or more per hour<\/a> on the worker platform. The single buyer-facing band on this search is $25 to $49 per hour<\/a>.<\/p>\n

Those two numbers are not directly comparable and this page is not going to pretend they are. They describe different markets, different engagement models and different work. An agency’s hourly rate carries quality assurance, tooling, project management, rework and margin inside it, and a rate paid directly to a contributor carries none of those. A generalist labelling task and a board-certified specialist reviewing clinical reasoning are not the same job priced differently; they are different jobs.<\/p>\n

What the comparison does give you is the shape of the purchase, which reading five vendor pages will not. It tells you that the labour input to a generalist programme is real but modest, that the labour input to a specialist programme is the dominant cost, and that any quote in a specialist domain which looks like a generalist quote deserves a question about who is actually doing the work.<\/p>\n

The specialist rows are where the two sides press hardest against each other, and they are exactly the rows a buyer commissioning medical, legal or scientific annotation needs to think about.<\/p>\n

Why this search shows you job listings<\/h2>\n

Three of the twelve results for this query are aimed at people looking for annotation work rather than companies looking to buy it: a Reddit thread in a work-from-home subreddit, a job board, and a contributor platform.<\/p>\n

That is not a search engine failure. It is one phrase describing both sides of the same transaction, and Google is showing both because both are legitimate readings of it.<\/p>\n

This page is written for the buyer. It is not a jobs page, it lists no openings and it makes no offer of work. The worker-side rate card appears here as evidence about what the underlying labour costs, which is information a buyer is entitled to and which no buyer-side page provides.<\/p>\n

It is worth noticing that no page on either side acknowledges the other exists. The vendor pages never mention what contributors earn. The contributor pages never mention what clients pay. The two halves of one market are documented separately, and the gap between them is where a buyer’s leverage lives.<\/p>\n

What to check before you shortlist anyone<\/h2>\n

Four filters, in this order, because the first two decide eligibility and the last two decide preference.<\/p>\n

Modality first.<\/strong> Image, video, text, audio, LiDAR and DICOM are genuinely different capabilities. A provider excellent at bounding boxes on street scenes is not thereby competent at segmenting medical imaging, and a provider strong on text ranking may have no computer vision practice at all. This eliminates most of any long list immediately.<\/p>\n

Compliance second.<\/strong> GDPR, HIPAA, SOC 2 and ISO 9001 are not badges, they are eligibility. If your data is regulated, a provider without the relevant certification cannot legally do your work, and no amount of price advantage changes that.<\/p>\n

Workforce composition third.<\/strong> Not size. A crowd of a hundred thousand generalists and a team of two hundred credentialed specialists are different products, and the one you need is decided by your data rather than by which number sounds more impressive.<\/p>\n

Managed service or tooling fourth.<\/strong> Decide before the first call whether you are buying labour or software, because the vendors who sell both will happily quote for whichever you seem to want.<\/p>\n

The ten providers<\/h2>\n

Each entry states what the provider is built around, who it suits and one real limitation. What no entry states is output quality, because nobody on this search publishes a measurement of it and we commissioned no work.<\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
The ten providers at a glance. All facts attributed to the page that published them, read 15 August 2026.<\/caption>\n
Provider<\/th>\nPrimary focus<\/th>\nBest suited to<\/th>\n<\/tr>\n<\/thead>\n
Scale AI<\/td>\nEnterprise annotation at scale<\/td>\nLarge programmes, with an ownership caveat<\/td>\n<\/tr>\n
Appen<\/td>\nMultilingual and volume collection<\/td>\nLanguage coverage at scale<\/td>\n<\/tr>\n
Surge AI<\/td>\nRLHF and LLM alignment<\/td>\nLabs aligning frontier models<\/td>\n<\/tr>\n
iMerit<\/td>\nRegulated, credentialed annotation<\/td>\nHealthcare, autonomous vehicles, geospatial<\/td>\n<\/tr>\n
Label Your Data<\/td>\nComputer vision and NLP services<\/td>\nMid-scale direct engagement<\/td>\n<\/tr>\n
Bright Data<\/td>\nWeb data collection<\/td>\nSourcing rather than labelling<\/td>\n<\/tr>\n
Cogito Tech<\/td>\nMultimodal with provenance<\/td>\nCompliance-led organisations<\/td>\n<\/tr>\n
Aya Data<\/td>\nFlexible, certified, tool-agnostic<\/td>\nMid-scale regulated work<\/td>\n<\/tr>\n
TELUS International AI<\/td>\nMultilingual and speech<\/td>\nEnterprise language programmes<\/td>\n<\/tr>\n
Innodata<\/td>\nEnterprise data services<\/td>\nLong-established services partner<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Scale AI<\/h3>\n

Lightly describes Scale AI as the largest enterprise annotation platform by revenue, and that positioning has held for several years across autonomous vehicles, defence and frontier model work.<\/p>\n

The fact a buyer needs before anything else is ownership. Lightly reports that Meta took a 49 percent stake in June 2025<\/a> and that chief executive Alexandr Wang moved across to lead Meta’s Superintelligence Labs. That is credited to Lightly and not independently verified here.<\/p>\n

Treated as gossip it is interesting. Treated as procurement it is decisive for a specific set of buyers. If you are building models that compete with Meta, you are weighing a supplier in which a competitor holds a substantial stake, and that raises questions about data handling, roadmap priority and confidentiality that exist entirely independently of how good the annotation is.<\/p>\n

For everyone else it may not matter at all, and saying so is part of reporting it honestly. A retailer labelling product images has no competitive relationship with Meta and no reason to care.<\/p>\n

Best for large enterprise programmes needing scale and established process. The limitations are the ownership question for a subset of buyers, and the absence of any published rate, which is true of almost every provider on this page.<\/p>\n

Appen<\/h3>\n

One of the oldest companies in this category, and Lightly credits it with a contributor base spanning 170 countries. Longevity matters more here than in most software categories, because annotation quality depends on process maturity rather than on product features.<\/p>\n

Appen’s strength is breadth<\/a>: language coverage, collection as well as labelling, and the operational experience to run programmes that involve tens of thousands of contributors without the whole thing degrading into inconsistency.<\/p>\n

That scale is also the limitation, and it is worth being direct about. A crowd model at that size makes consistency a management problem rather than a hiring problem. The quality you receive depends heavily on how well the task is specified, how the gold-standard set is built, and how disagreement between annotators is resolved. Those are things you influence, which means a poorly specified brief produces poor output regardless of the vendor’s reputation.<\/p>\n

Best for multilingual work, large-volume collection, and programmes where geographic spread is the requirement. The limitation is that scale demands specification discipline from the buyer, and no rate is published anywhere on its pages or in the comparisons that name it.<\/p>\n

Surge AI<\/h3>\n

Surge AI specialises in reinforcement learning from human feedback and language model alignment, and Lightly reports it serving OpenAI, Google, Anthropic and Microsoft. That client list is credited to Lightly and explicitly not verified here.<\/p>\n

Understanding what this work is explains why the category changed. RLHF is not labelling in the traditional sense. It is preference ranking, where a person compares two model outputs and judges which is better, and red-teaming, where a person tries to make a model behave badly so the failure can be trained out. Both require judgement and domain knowledge rather than throughput.<\/p>\n

That shift is visible in the worker-side rate card at the top of this page. Generalist labelling sits at $25 to $30 an hour; the domains where you need a lawyer, a physician or a mathematician sit at $50 to $100 or more. Alignment work draws almost entirely from the second group, which is why it is the expensive end of the category and why the providers doing it look nothing like the providers doing bounding boxes.<\/p>\n

Best for organisations aligning or evaluating frontier models, and for anyone whose evaluation needs expert judgement rather than volume. The limitation is that this is a fundamentally different and more expensive product than bulk annotation, and buying it for a bulk task wastes money.<\/p>\n

iMerit<\/h3>\n

Founded in 2012 and based in San Jose, iMerit is the clearest illustration on this page of why workforce composition beats workforce size as a selection criterion.<\/p>\n

Lightly credits a workforce of more than 7,000 that includes credentialed domain specialists<\/a>: board-certified pathologists for medical image annotation, and sensor-fusion specialists for autonomous vehicle work. Those are not annotators trained on a medical task. They are practitioners of the discipline doing annotation within it.<\/p>\n

For regulated work that distinction is the entire purchase. A pathology annotation that a regulator or a clinical partner will accept has to be produced by somebody whose judgement is defensible, and “we trained our team on medical imaging” is not a defence. This is also why audit-readiness rather than throughput is the metric that matters in these domains.<\/p>\n

The same logic applies to autonomous vehicles, where sensor fusion across camera, radar and LiDAR is a specialism rather than a labelling task, and errors have consequences that a generalist workforce is not equipped to reason about.<\/p>\n

Best for healthcare, autonomous vehicles and geospatial work where annotation must survive scrutiny. The limitation is straightforward: credentialed specialists are the expensive end of the market, and using them for work that does not need them is an expensive mistake.<\/p>\n

Label Your Data<\/h3>\n

Label Your Data<\/a> is an annotation company covering computer vision and natural language work, and it ranks on this query with its own home page rather than with a comparison article, which is worth noting as a difference in approach from most of this set.<\/p>\n

What it publishes about its own work is more concrete than most vendor pages manage: polygonal annotation and drone data appear among its named case examples, which tells a prospective buyer something specific about the shape of projects it takes rather than a generic capability list.<\/p>\n

Its page also carries star ratings and review counts from third-party directories. Those are excluded from this article entirely, along with every other rating on this search, because a rating with no date and no methodology attached is not evidence a reader can act on, and repeating it would lend it a credibility this page cannot verify.<\/p>\n

Best for mid-scale projects where direct engagement with the provider matters more than the procurement machinery of an enterprise vendor. That is a real advantage for a team that wants to talk to the people running the work rather than to an account manager.<\/p>\n

The limitation is the one shared across this entire category: no rate is published, so the only way to learn what it costs is to ask.<\/p>\n

Bright Data<\/h3>\n

Bright Data<\/a> solves a different problem from everyone else on this page and the distinction is one buyers routinely miss.<\/p>\n

It supplies web data collection: datasets gathered at scale from public sources, delivered as raw material. It does not label that material for you. Collection and annotation are two separate purchases, and a team that needs both is either buying from two suppliers or from one that does one of them less well.<\/p>\n

Knowing which problem you actually have is the useful thing here. A team with plenty of internal data and no labelling capacity needs an annotation provider. A team with a well-drilled labelling process and nothing to point it at needs a collection provider. A team that has neither should solve sourcing first, because annotating the wrong data carefully is the most expensive failure mode in this whole category.<\/p>\n

Web collection also carries its own considerations that annotation does not: the legality and terms-of-service position of the sources, the provenance documentation you will need if the resulting model is ever scrutinised, and the freshness of the data relative to your use case.<\/p>\n

Best for teams needing volume source data. The limitation is definitional: it does not solve the labelling problem, only the problem of having something to label.<\/p>\n

Cogito Tech<\/h3>\n

Cogito Tech works across images, video, audio, text, DICOM and LiDAR, which Lightly credits as one of the broader modality ranges in the category, and it publishes a provenance framework it calls DataSum.<\/p>\n

Provenance documentation deserves more attention than most buyers currently give it. As data lineage moves from a nice-to-have into contracts and regulation, being able to demonstrate where every training example came from, who labelled it, under what instructions and with what review, stops being an internal quality matter and becomes an external one.<\/p>\n

A provider that has built a named framework around this has at least thought about the problem structurally, which is more than can be said for a vendor whose answer is that it keeps records. Whether the framework does what it claims is a question for a pilot, not for a comparison article.<\/p>\n

The breadth across modalities is the other selling point and also the thing to interrogate. Breadth is a claim about capability across many disciplines, and the honest test is to ask about depth in the one modality you actually need rather than being reassured by the length of the list.<\/p>\n

Best for compliance-focused organisations working across several data types. The limitation is that broad claims need narrow verification.<\/p>\n

Aya Data<\/h3>\n

Aya Data offers flexible, tool-agnostic annotation across computer vision, natural language and agricultural AI, and Lightly credits it with holding GDPR, HIPAA, SOC 2 and ISO 9001.<\/p>\n

That is the most complete certification set named anywhere on this search, and for a mid-scale buyer it is the reason to put Aya on a shortlist. Certification is an eligibility question rather than a quality one, but it is an eligibility question that removes most providers from consideration for regulated work, and holding all four means a single provider can cover privacy, health data, security controls and quality process without a patchwork.<\/p>\n

Tool-agnostic is the other useful attribute and an underrated one. A provider tied to its own platform creates a dependency: your labels live in their system, your process is shaped by their interface, and moving is expensive. A provider that will work in whatever tool you already run leaves that decision with you.<\/p>\n

The agricultural AI specialism is unusual enough to mention, because it is a genuinely awkward domain, involving field imagery, seasonal variation and crop-specific judgement that general providers handle poorly.<\/p>\n

Best for mid-scale projects needing certification without enterprise pricing and process. The limitation is size: for very large programmes the enterprise providers have operational depth Aya does not.<\/p>\n

TELUS International AI<\/h3>\n

TELUS International AI absorbed Lionbridge AI and offers managed labelling focused on multilingual tasks, speech recognition evaluation and large structured training programmes, per Lightly. Encord notes its Ground Truth studio offering three platforms as part of a managed service<\/a>.<\/p>\n

It also appears in the only buyer-generated shortlist visible on this search. A thread on the DeepLearning.AI community, where somebody asked for recommendations to annotate 100,000 images, produced TaskUs, TELUS International, Appen, Scale AI, iMerit and Damco. That list was written by practitioners rather than marketers, which makes it the most interesting artefact on the search, and it could not be extracted for this article, so it is reported from its summary alone.<\/p>\n

The multilingual and speech focus is the differentiator. Speech recognition evaluation in particular is a specialism: it needs native speakers, accent coverage and a consistent judgement standard across languages, which is an operational problem more than a technical one.<\/p>\n

Best for enterprise language and speech programmes where coverage across many languages is the requirement. The limitation is an enterprise engagement model, with the procurement cycle that implies, and no published rate.<\/p>\n

Innodata<\/h3>\n

Innodata provides enterprise AI data services spanning large-scale annotation, content processing and AI training support, and it is long established relative to most of this category.<\/p>\n

What makes it worth including is not only its capability but where it appears. Innodata surfaces on the worker-facing side of this search as well as the buyer-facing side, listed among companies people apply to for annotation work. That makes it a useful illustration of the two-audience point rather than a separate finding: the same company is a supplier to one reader and an employer to another, and both found it through the same query.<\/p>\n

Its positioning is the traditional services model: an established organisation with process, contracts and account management, aimed at enterprises that want a partner rather than a platform. For buyers in regulated or content-heavy industries, that shape is often easier to procure than a newer vendor, whatever the relative technical merits.<\/p>\n

Best for enterprises wanting a long-established services partner with content processing alongside annotation. The limitation is the one this entire page keeps returning to: you cannot compare it on price, because like almost everyone here it does not publish one.<\/p>\n

Where the directories fit<\/h2>\n

One directory ranks for this query, and it is simultaneously the most useful and the most carefully-handled source on this page.<\/p>\n

GoodFirms is useful because it publishes the only buyer-facing rate anywhere on the search: its top-ranked firm listed at $25 to $49 per hour, with a founding year and a location. In a category where nobody publishes prices, one real band is worth a great deal.<\/p>\n

It needs careful handling because that same page discloses that it carries sponsored placements. The disclosure is to its credit; plenty of directories do not make one. But it means the ordering is a mix of assessment and commercial arrangement, and the two are not separable from the outside.<\/p>\n

So this page uses the rate and does not reproduce the ranking. Repeating a sponsored order inside an editorial article launders a commercial arrangement into an endorsement, and a reader has no way to tell which parts of the list they are reading.<\/p>\n

The general principle is worth carrying to any directory in any category: take the structured data, which is usually accurate, and leave the ordering, which usually is not what it appears to be.<\/p>\n

What the tables cannot tell you<\/h2>\n

Output quality, which is the only thing that actually matters, and which not one page on this search puts a verifiable number against.<\/p>\n

Every accuracy figure on this search is a vendor’s claim about its own work with no method published: no test set named, no inter-annotator agreement reported, no independent verification. We commissioned no annotation and evaluated no output, so this page offers no ranking on quality and would be inventing one if it did.<\/p>\n

No independent benchmark of these providers surfaced anywhere in the research, which is itself a finding about the category. In a market this large and this well funded, the absence of any third-party evaluation is striking.<\/p>\n

What to do instead is concrete. Run a paid pilot on your own data, with a gold-standard set that you build and keep to yourself, scored by you against criteria fixed before the work starts. Send the same pilot to two or three providers. That costs a few thousand dollars and it is the only evidence about quality that will ever be specific to your problem.<\/p>\n

How this guide was built<\/h2>\n

Twelve results were reviewed and eight were extracted in full on 15 August 2026, with no extraction failures. Keyword research was run the same day against Google’s autocomplete, returning 400 queries. That source publishes no search volume and none is quoted here.<\/p>\n

Three of those twelve results serve workers rather than buyers, and that is reported as a fact about the search rather than filtered out, because the worker-side rate card is the only substantial pricing evidence available in the category.<\/p>\n

Every buyer-side list on this search is published by a company selling in it. Lightly supplies more of the detail on this page than any other source and is credited throughout, including where it ranks its own product first, because publishing checkable specifics about competitors is more useful than publishing nothing. GoodFirms discloses sponsored placements and its ordering is deliberately not reproduced here.<\/p>\n

Two forum results could not be extracted, including the DeepLearning.AI thread carrying the only buyer-generated shortlist on the search. That is a real gap: practitioners recommending providers to each other is better evidence than any vendor page, and this article has it only in summary.<\/p>\n

If you work in this category and something here is out of date or wrong, the editorial contact page<\/a> is the fastest route to a correction.<\/p>\n

Frequently asked questions<\/h2>\n

These are the questions buyers actually search, taken from Google’s own suggestions and from the sections the ranking pages publish.<\/p>\n

What is data annotation?<\/h3>\n

Labelling raw images, video, text, audio or sensor data so a model can learn from it: boxes around objects, tags on entities, transcriptions of speech, or rankings of one model output against another. The label is the signal; the raw file teaches nothing on its own.<\/p>\n

How much does data annotation cost?<\/h3>\n

Almost nobody publishes a rate. The one buyer-facing band on this search is $25 to $49 per hour from a directory listing. On the worker side, generalist work pays $25 to $30 or more per hour and specialist domains $50 to $100 or more, which tells you where the cost sits.<\/p>\n

Is data labeling different from data annotation?<\/h3>\n

Not meaningfully. This entire result set uses the two terms interchangeably, sometimes within a single sentence. A page insisting on a sharp distinction is usually making a positioning argument rather than a technical one.<\/p>\n

Is annotation done by humans or by AI?<\/h3>\n

Both, in combination. Automated pre-labelling handles the straightforward cases and humans handle the ambiguous ones and the review. The proportion is what the price reflects, and the domains needing most human judgement are the ones paying $50 to $100 an hour on the worker side.<\/p>\n

What certifications should a data annotation vendor have?<\/h3>\n

It depends on your data, but GDPR, HIPAA, SOC 2 and ISO 9001 are the set named on this search. These are eligibility rather than quality: for regulated data, a provider without the relevant certification cannot do your work at any price.<\/p>\n

Which data annotation company is best?<\/h3>\n

This page cannot tell you and no page on this search honestly can. We commissioned no work and evaluated no output, and no independent benchmark of these providers exists. Run a paid pilot on your own data against your own gold-standard set instead.<\/p>\n","protected":false},"excerpt":{"rendered":"

Search for data annotation companies and Google returns two different markets at once: the companies selling annotation, and the people doing it. The hourly rate is published only on the second group’s pages, and it is the closest thing to a price anybody publishes in this category. Every buyer-facing list ranking for this question is […]<\/p>\n","protected":false},"author":1,"featured_media":396,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-380","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog"],"_links":{"self":[{"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/posts\/380","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/comments?post=380"}],"version-history":[{"count":1,"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/posts\/380\/revisions"}],"predecessor-version":[{"id":381,"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/posts\/380\/revisions\/381"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/media\/396"}],"wp:attachment":[{"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/media?parent=380"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/categories?post=380"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/texttoolz.com\/blog\/wp-json\/wp\/v2\/tags?post=380"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}