What an AI consultant does, and which AI role you actually need.

What does an AI consultant do?

You hire an AI consultant for one of two jobs, or for both: to decide what you build and why, or to build it and ship it. Ask for a written scope that names which job you are buying. Each one needs different people. Your strategist ranks your use cases by value and feasibility, sets your governance guardrails, and gets your leadership aligned. Your builder takes on your data pipelines, your retrieval architecture and your model selection, and after that your evaluation, security and integration. Some consultants claim both jobs (a few can do both well, though it's rare). If one does, ask which they would rather do for you.

Your scoped assessment project runs two to four weeks, according to Alice Labs, and you should come out of it with a ranked list of your use cases that you can read in one sitting. RAND research, as cited by Iternal, finds that the most common cause of AI failure is leaders misunderstanding or miscommunicating the problem to be solved. So can your leaders state your problem in one sentence? Does your scope say it? If not, settle the problem statement with them before you pay for your build. Skip that step and your money goes missing, and no one will be able to say where it went.

Your free assessment does not replace that conversation. It reads only your public pages, so you see what a crawler sees, not what your team needs. But it still hands you something useful at no cost: the crawler's view of your site. Run it on your own site.

Which AI roles and services exist, and which one do you need?

Match the role to the decision in front of you: a strategist if you do not know what to build, a builder if you do. Check the verbs in your last two AI conversations. Did you or your team say choose, prioritize or govern? Then you need strategy. If you said connect, deploy or monitor, you need an implementation consultant, who writes code and builds pipelines. Your verbs give you away, and they rarely lie.

Your options span generative and LLM projects, agentic systems, workflow automation and integration, retrieval and data, and application development and operations. CGI also lists predictive analytics, natural language processing, computer vision, recommendation engines and intelligent document processing as common builds, and you may spot yours there. Below, each role comes with the situation in which you would hire it. Match it to your plan. And expect overlap: most projects touch more than one role, so yours may too.

You likely need one or two of these, not all of them. If none fits your situation, you may not need a consultant yet. Run the verb check to see which applies.

When do you need an AI consultant rather than a hire or an agency?

You need a consultant when your work is a decision or a bounded build. You need a hire when your work is daily and permanent, and an agency when you want delivery capacity you do not want to employ. Put your budget beside two numbers. An AI consultant on staff costs a median around $207,000, according to MentorCruise citing Salary.com, while a fractional advisor runs $5,000 to $15,000 a month at 10 to 20 hours a week. That gap (large, and not subtle) is the case for matching your arrangement to your work.

Choose a consultant over a hire when you need a scoped answer in two to four weeks, such as which of your use cases to fund. If your system needs an owner every day after launch, hire someone. I consult on AI for websites and web estates, and I tell people plainly when they need a hire instead. An agency fits a fixed production job. Whoever you pick, ask who owns your system after handover.

What should an AI consultant assess first on your website?

Your first check is whether AI crawlers can reach your pages and whether your pages hold passages worth quoting. Open your robots.txt and see whether it names OAI-SearchBot, GPTBot and PerplexityBot. OpenAI treats OAI-SearchBot and GPTBot as independent settings, so you can appear in search results while you disallow GPTBot to keep your content out of training. That choice is yours, and it is a real one.

Perplexity recommends that you allow PerplexityBot, which links to your site in its results and is not used to train foundation models. Two fetchers that act on a user's request, ChatGPT-User and Perplexity-User, may ignore your robots.txt because a person asked for your page. On your key pages, check that each answer sits in a short quotable passage and that the page carries structured data. Be patient. OpenAI says search results can take about 24 hours to adjust after you update robots.txt, so do not judge your change an hour after you make it.

Start with the AI readiness pillar of the assessment, which checks crawler access, quotable passages and structured data across your sampled pages. Run it on your site. You will see where you stand on all three.

How much does an AI consultant cost?

Expect to pay about $150 to $300 an hour for a typical AI consultant, and you will see $80 to $600 across the full market, per Groovy Web. A solo expert charges $80 to $200, a boutique consultancy $150 to $300 and a Big Four firm $300 to $600. Sources disagree at the edges, so check them. AIDOLS puts Big Four firms at $500 to $1,000 or more.

Project quotes vary more. AIDOLS lists $25K to $75K for an assessment, $50K to $250K for a proof of concept and $100K to $500K for a single use case. If you go boutique, Groovy Web's range is $50K to $300K over one to four months. Need only an assessment? Iternal lists $7K to $35K, lower than AIDOLS. A fractional advisor runs $5,000 to $15,000 a month. Put the rate card and the scope in writing. Compare quotes on one deliverable, because quotes for different deliverables tell you nothing.

How do you judge whether an AI consultant's advice is sound?

You can judge an AI consultant's advice by asking for evidence you can check. Look for a written problem statement you can read, a measure of success you can test, and a plain account of what could fail on your site. And put four questions to your consultant on your first call. What problem are you solving for your site? How will you evaluate the result? Who owns the system after launch, you or them? And what would make you advise against building anything at all (a good consultant has an answer ready)?

If your consultant skips that first question, treat it as a warning sign and press for an answer. RAND, as cited by Iternal, finds that leaders misunderstanding the problem is the most common cause of AI failure. So check your own brief too. Be wary when a consultant names tools before your problem, promises you outcomes from AI, or quotes an hourly rate with no scope attached. Ask to see how your consultant evaluated the last project, and then judge what you are shown. But keep in mind that the assessment reads your public pages only, so your score is a first read, not a verdict.

What does AI readiness mean for your website specifically?

For your website, AI readiness means AI systems can reach your pages, find a passage worth quoting, and read what each page is about from your structured data. Open your homepage source and look in the markup for structured data. Read your first heading and the paragraph under it, then ask whether you could quote it on its own. If you cannot, a model will struggle with your page too.

You may know a different definition. Monterail and N-iX describe readiness as data, infrastructure, talent and governance, assessed over two to six weeks. N-iX cites IBM's 2025 study finding only 26% of organizations feel confident their data can support AI initiatives. Could you say the same of your data? Your public pages cannot answer that. The assessment was never meant to.

You can run the free assessment in about two minutes. The AI readiness pillar carries 10% of your score. You enter your work email once, so it can send you your report link. That is all it asks of you.

Every AI role, in one list.

Questions, answered.

Your next read.

1. 01AI SEO auditCrawl access, extractable passages, schema, llms.txt.
2. 02Website consultantWhat a consultant checks first, and what it costs.
3. 03Build a website with Claude: what you get, and what to check before launchWhat Ships · Claude