An SEO audit agency is paid to end an argument. It establishes what is actually true about your search performance, forms a judgment about what that costs you, and then commits to an order of work in writing. Most firms manage the first part well, the second part unevenly, and avoid the third.

Establishing what is true. Not what a dashboard reports and not what the previous vendor claimed. Which addresses are indexed, which of them earn anything, whether the conversion numbers can be trusted at all, how much demand genuinely exists in your category, who currently holds the positions you want and by what method, and how your business gets described by sources you do not control.

Forming a judgment. A finding only becomes useful once somebody weighs it against your margins, your value per enquiry, your release cycle and your appetite for disruption. Without that weighing you have a fact. With it you have a decision worth money.

Committing to an order. Search work compounds and it also has prerequisites, so doing six correct things in the wrong sequence can waste two quarters. An audit that hands over a flat list has declined the part of the job that requires being wrong out loud in front of a client.

The studio opened in 2018 and has since delivered more than 500 projects for roughly 150 brands, with four Awwwards and a Webby against the design work. Independent verification sits at a 5.0 Clutch score assembled from 64 or more client interviews we did not run. Leads have risen 250% on average across past engagements, which is a record and not a projection, since it averages accounts that began in very unlike conditions, several of them running past two years, and none of them yours. The senior team behind the diagnosis is the same team that implements what it recommends, which is unusual and it is the reason a finding here reaches production instead of a slide. Reading server logs, modelling demand and reviewing conversion tracking are three different skills, so the roles we hire into are deliberately mixed, and our Toronto studio holds all three in one room.

Where this sits commercially, it is the diagnostic that precedes anything ongoing in the search practice as a whole, and buying it does not obligate you to buy what it recommends.

Eight examinations, run as one investigation, because the value is in reading them against each other instead of one at a time. A technical fault means something different once you know whether demand sits behind the pages it affects.

1. Technical audit. Whether the site can be reached, rendered, indexed and trusted. Index coverage, what a bot receives compared with what a browser assembles, depth and internal links, duplicate addresses competing for one position, redirects, whether the phone version carries everything the desktop one does, and performance on real devices. Every item is judged by what it costs you and never by the label a tool attached to it, the standard throughout technical SEO.
2. Content and on-page audit. Every page inventoried against three questions, meaning what it produces today, what it could produce, and whether it answers the query it targets. Then the overlaps where your own pages compete, the pages losing ground, and the titles, headings, entity coverage and linking on everything worth keeping. It hands off into SEO content strategy.
3. Backlink and authority audit. Your referring domains against your competitors', with the method each used to build it. Which links carry weight, which were lost, what the wording of the links pointing at you says about you, and whether anything in the profile is a liability. This is the diagnostic half of link building and digital PR.
4. Competitor audit. Which companies actually own your demand instead of the ones your team worries about, the page types they win with, where their coverage is thin, and which of their advantages you could copy inside a year. Naming the ones you cannot matters as much.
5. Search demand and keyword opportunity analysis. What there is to win, grouped by what the searcher wants, with commercial value, difficulty, seasonality and direction of travel attached, including the phrases you are ranking for without meaning to, usually the cheapest gains available.
6. Analytics and tracking verification. Whether the numbers everything else is judged against describe reality. It runs first and has its own answer below.
7. AI visibility audit. What assistants say about your category, who gets named instead of you, which sources those answers are built from, and how you are described when you appear. Several of the largest crawlers fetch a site's script files and never execute them, so anything the browser assembles is absent from what they collect. Remediation belongs to AI search optimization.
8. The prioritized roadmap. The product of the other seven and the only part with decisions in it.

How heavily each area is weighted comes out of the first week. A single-discipline read is a legitimate smaller purchase, and on plenty of sites the honest scope is three of the eight.

A crawler returns findings. An audit returns decisions, each carrying a business case and a position in a sequence. The gap between those two things is the entire fee, and it is worth being precise about where the gap sits.

Give the tools their due first. Modern crawlers are extraordinary. They will inventory a million addresses overnight, catch every broken link, every duplicate title, every response code out of place, and they do it more completely and more cheaply than any person could. We run several of them on every engagement and we would be worse at this without them. What they cannot do is the part you are actually buying.

• They have no idea what a page is worth. Severity is assigned from a rule written to apply to every website in existence, so a missing meta description on a dormant blog post and an instruction telling search engines to drop your highest-converting service page can arrive labelled the same way.
• They cannot see your business. Margin, value per enquiry, sales cycle, stock position, which twelve pages produce the revenue. None of it is visible to software pointed at a domain.
• They cannot see demand. A page can pass every check in the report and target a phrase nobody types. A tool has no mechanism for telling you that the correct answer is to delete it.
• They cannot read a competitor's advantage. A rival winning on twelve years of trade coverage and a rival winning on better category pages need opposite responses, and no crawl distinguishes them.
• They have no view on order. Publishing thirty pages into a template that renders nothing for a bot spends the same budget twice. Sequence is most of the return in search and it is the one thing no export contains.
• They cannot decide. A findings list transfers the hardest judgment back to the person who paid to avoid making it.

Three things turn a finding into a decision. The first is an estimate of what acting on it is likely to return, with the assumptions printed beside it. The second is a named owner type, so you know whether the item needs a writer, a marketer or the people who deploy it. The third is a place in the order, including the items we recommend never touching. On a large product catalogue a crawl will hand you two hundred thousand filter addresses and no opinion about which of them deserve to exist, and that opinion is the work.

Three different questions, producing three different documents, and buying the wrong one is the most common way an audit budget disappears. They overlap in the middle and they answer to different standards of evidence.

The SEO audit asks whether you get found and chosen in search. Its scope is everything upstream of the click, meaning eligibility, coverage, authority, demand, competitors and the answers assistants give about your category. It is the right purchase when visibility is falling, never arrived, or sits well below what your competitors hold. A technical-only version of it is a smaller and cheaper scope, and where the evidence points that way the work narrows instead of running wide.

The marketing audit asks whether the plan holds together. It is wider and shallower by design, covering audience, message, offer, channel mix, spend allocation and measurement across everything you run. Its classic case is the business where every channel reports acceptable numbers and the total refuses to move, which is a problem living in the joints between channels instead of inside any one of them. That is a marketing audit.

The UX audit asks whether people can finish what they came to do. It is deep and narrow, taking the two or three paths where your revenue actually happens and walking each one end to end against observed behaviour, with severity attached to every point where confidence drains away. It is the right purchase when the traffic arrives and nothing happens, and that is a UX audit.

A rough diagnostic follows, and it is usually enough to self-select on a first call. If your important pages cannot be found, start with search. If they are found and people leave without acting, start with usability. If they are found, people convert, and the business is still flat, the question is strategic and belongs in the wider review.

The honest complication is that all three read your analytics, all three will notice a slow page, and the boundaries between them are invisible to a customer. So scope gets set to your question and not to a service name. We have sold this audit and concluded in week three that search was working and the destination was the problem, which meant handing a share of the budget to a discipline with a different name on it. That happens often enough to be worth saying before you buy.

An SEO audit is sold as a fixed-scope engagement with a defined end and never as a retainer, and most run three to five weeks from the day access lands to the working session. The deliverable is finite, which means you should be able to purchase it without agreeing to anything that follows.

Days one to five. Access and baseline. Analytics, Search Console, the content system, rank tracking history, whatever the CRM holds on enquiry sources, and server or delivery-network logs where obtainable. The starting numbers get recorded before anything is judged, so the same measurements can be repeated later.

Weeks one to three. Examination. The eight areas, worked in the order set out further down this page. Findings get evidence attached as they are produced instead of reconstructed at the end.

Weeks three to four. Prioritization. Every surviving finding gets tested against demand, against your competitors, and against your own release cycle and staffing. A long list becomes a short sequence here.

Weeks four to five. Delivery. A working session with the people who can approve things, then the document. Reading the awkward finding out loud to a room is the only reliable defence against it becoming a file nobody opens.

What moves the number.

• Template count, not page count. Twelve templates behind a hundred thousand addresses is a smaller job than four hundred pages built by nine people over six years.
• How many of the eight areas are in scope, and whether the competitor and demand work needs primary research.
• The condition of the measurement. Where the tracking cannot be trusted, repairing enough of it to judge anything becomes part of the scope, quoted openly instead of absorbed.
• Whether logs are obtainable at all, which depends on your stack.
• Markets, languages and locations, since each multiplies the demand and competitor work.
• Estate size, since the same questions across a very large property get scoped as enterprise SEO.

Two commercial points. Where an audit leads into work with us, the fee comes off the first phase, and the credit carries no expiry, because a deadline on a credit is a discount pretending to be urgency. Ongoing work that follows carries a six-month minimum, set out in the ongoing program terms, which stops a client paying for the build-up and leaving before the payback.

On results, the audit itself produces none. Execution does, and real movement takes from four months to a year to reach your pipeline. Any firm promising first-page positions inside thirty days is describing something that was already broken or chasing terms nobody types. Send us the domain and you will get a scope and a straight read on whether it is worth buying.

The analysis is rarely the problem. The document stops one step short of a decision, and the step it skips is the only one a client genuinely cannot take without help. A reader who finishes agreeing that all one hundred and forty findings matter has been given a description of their website and no way to start.

Three tests will tell you whether an audit you already own is a decision document. Does any single item carry a number for what fixing it should be worth. Does the list have an order with reasons attached to the order. And does it name anything you should deliberately leave alone. A document failing all three has cost you money and bought you a feeling of progress.

Five ways they stop short, in rough order of how often we see them.

1. Severity stands in for revenue. The ranking comes from the tool's own labels, so a site-wide cosmetic warning outranks a single blocked page that happens to be where the enquiries come from.
2. There is no sequence. Search has prerequisites. Commissioning an editorial plan for a site whose templates deliver nothing to a crawler, or buying off-site authority work for pages that are not indexed, spends the budget on a result that cannot arrive yet.
3. Nothing carries a value estimate. An item with no number beside it cannot be approved, so it waits for a budget conversation that never gets booked.
4. Nobody says what to ignore. Exclusions are the only part of an audit where the author can be publicly wrong, which is exactly why they get left out. An audit with no skip list has taken no risk on your behalf.
5. It was written by somebody who will never implement it. A recommendation that needs a release from a team with a locked roadmap and a different set of priorities is a wish, and the person who wrote it faces no consequence for that. Where an item needs a development slot, the document has to say whose, and when.

There is a sixth reason and it is less comfortable. Sometimes the audit was commissioned to support a decision already taken, or to win an internal argument in a particular direction, and the finding that disagrees gets quietly set aside. No amount of formatting fixes that. What helps a little is delivering the finding live, to the people who can act, with the disagreement happening in the room instead of in silence three weeks later.

Four situations, and in every one of them you should keep the money. Saying so costs us work and it is the only version of this that survives contact with a sophisticated buyer.

1. You can already point at the problem and prove it. Where somebody internally can name the cause and produce the evidence, an audit is a receipt for something you already own. Buy the repair. Tell us on the first call what you already know, and we will say whether the evidence is strong enough to act on without a second opinion.
2. A rebuild is funded and lands inside two quarters. The useful purchase then is a review of the architecture arriving, held while templates, addresses, rendering approach and the measurement plan are still open for argument. Auditing a structure with a deletion date on it buys a document that expires on launch day. That review sits inside a website redesign engagement and it produces specifications instead of findings.
3. You have a competent audit from the last twelve months that nobody executed. A second document does not create capacity, and the constraint was never analysis. Buy execution, or buy two hours of somebody sitting with the document you already have and turning it into a sequence with owners on it. A wider marketing read is the cheaper way to get that.
4. The constraint is not search. Where your important pages already rank and the traffic leaves without acting, a search diagnosis will tell you nothing you can use, and a usability review is the correct spend. Where almost nobody types what you sell, there is no demand waiting to be captured and it has to be manufactured first, which is a brand and paid media conversation.

The purchase that looks like this one and is not. A crawl report with commentary, sold cheaply and quickly, is a genuine product and a different one. It has value where you have technical staff who will act on it and no interest in strategy. It is worthless where the question is what to do first.

A smaller version exists for almost every situation. One discipline instead of eight, or a two-week read of the technical layer and the measurement only, or a scoped conversation on the single decision in front of you. And the outcome is allowed to be do nothing. On sites in reasonable condition in categories with limited demand, the honest finding is that the site is not the ceiling, and we will write that down and hand it over.

Measurement first, then eligibility, then merit, then demand, then the outside world. Running that order backwards is how an audit reaches confident conclusions from numbers nobody checked, which is the most expensive mistake available in this work.

1. Analytics and conversion tracking. Verified before a single judgment gets made, for reasons that fill the next answer.
2. Inventory and access. A full crawl, the sitemaps, the index counts, a log extract where one exists, and query and landing page exports going back far enough to show a trend.
3. Eligibility. Whether pages can be discovered, requested, rendered and kept. Anything failing here invalidates every quality judgment downstream, so it gets settled before anybody reads a word of copy.
4. Architecture and internal linking. Depth from the homepage, how many links point at what, and whether the structure states which pages you consider important. Most sites say something different from what their owners believe.
5. Content and on-page, page by page. What each page earns, what it targets, whether it matches the intent behind that query, and where it overlaps a neighbour. This is where editorial planning gets its brief.
6. Search demand. What exists to be won, what the searcher actually wants in each case, when in the year it happens, and whether the volume is growing or draining away. Seasonality changes conclusions more than people expect, and in property and development marketing a quarter read in isolation will mislead you every time.
7. Competitors. Who holds the positions, on what page types, and by what method. Read after demand, because a rival's win only matters where the demand behind it is real.
8. Authority. Your profile and theirs, once we know which pages and which terms the authority would be aimed at.
9. AI answers. The prompt set built and run, sources recorded, crawler access confirmed. Last, because it depends on almost everything above it.

The order exists because each step changes how the next one reads. A thin page is a nothing finding when the term has no demand behind it and an urgent one when a competitor is earning from that term today. A weak link profile is a priority in a contested category and a distraction in a category where nobody has any links at all. Where a business operates across many physical markets, step six splits by market before anything else happens, which is the logic behind local SEO work.

One exception. Where the numbers collapsed on a specific date somebody in the building can still name, we start there and work outward, because a dated collapse usually has one cause and finding it is worth more than a complete inventory. That runs alongside the technical stream instead of waiting for it.

Because every recommendation in the document is an argument about money, and an argument about money built on measurement nobody checked is a guess with a chart behind it. On most sites we inherit, something in the tracking is wrong, and usually in a direction that flatters or condemns the channel we were hired to assess.

What gets checked, and what turns up.

• What the conversion events actually count. Form views recorded as submissions. A thank-you page firing again on refresh. One enquiry counted twice by a tag manager and a platform tag running side by side. Each of those makes a page look better than it is.
• Whether the events arrive at all. A form built by a third party, posting to a subdomain, frequently records nothing in the property everybody reports from.
• Channel grouping and referrers. Organic sessions landing in direct, self-referrals from a payment step, campaign parameters overwriting the original source, and traffic from assistant platforms arriving with no referrer and getting filed as direct. That last one understates how the assistants read you on nearly every account.
• Cross-domain continuity. Where the journey crosses from a marketing site into a booking system, a checkout or a logged-in application, the session usually breaks at the boundary, which matters most in software businesses where the product sits on a subdomain.
• Consent and its effect on volume. A consent banner introduced part-way through the year removes a share of recorded sessions, so comparing the months either side of it produces a decline that never happened. Requirements differ by jurisdiction and belong with your own counsel.
• Calls and outcomes that happen off the site. Where the enquiry is a phone call and calls are uncounted, an entire channel looks like a failure. Where the CRM knows which leads closed and the analytics does not, nobody can tell a cheap lead from a good one.
• Bots, internal traffic and duplicated properties. Office visits, staging traffic, and two properties on one site disagreeing by thirty percent.

Then a decision gets made in the open. Either the repair becomes part of the scope, quoted separately, or every downstream number carries a label saying it is directional and why. What does not happen is a roadmap built on numbers we know are wrong.

Sometimes the whole finding is here. We have delivered audits whose headline was that organic had been performing acceptably for a year and the reporting had been hiding it, which is an uncomfortable thing to tell a marketing director and cheaper than the alternative. Reporting from that point sits on leads, qualified pipeline and revenue and never on impressions, and the same verified events are what the paid side optimizes against.

Every surviving finding gets scored on three things, and the third reorders the list more than the other two combined. What it is likely worth, what it costs to implement, and whether your organization can ship it in the next two quarters.

What it is worth. Modelled from the demand behind the affected pages, where they sit now, a realistic click share for a reachable position, the conversion rate those pages already record, and your own value per enquiry. All five are assumptions, so all five get printed where you can attack them. A range with visible assumptions beats a confident figure with none, and we would rather be argued with than believed.

What it costs to implement. Measured in whose calendar it needs instead of in hours. A setting somebody changes this afternoon, a brief a writer picks up, a template change waiting on a release, and an architectural rebuild needing a quarter of engineering are four separate currencies, and only the last competes with your product roadmap.

Whether you can actually ship it. Asked during scoping instead of discovered at delivery. A recommendation you have no route to executing is a description of a problem wearing a verb. A smaller fix a marketer publishes on Thursday frequently outranks a larger one needing a team whose year is already committed.

Dependencies, which override all three scores. This is where search diverges from a general ranking exercise. Some items are worth nothing alone and unlock everything after them, so they climb above better-scoring work. Rendering before content. Indexation before authority. Measurement before any of it. The document marks those chains, so a client reordering the list for good internal reasons can see which moves break one.

What happens at the bottom matters as much as the top. Low value against high effort earns a written recommendation to leave it alone, with the reasoning attached, so nobody revisits it next year. Low value against low effort gets grouped into one cleanup batch, shipped alongside normal releases instead of occupying the front page.

The honest ceiling. These are estimates built four weeks into knowing your business, and the first ninety days of execution will move some. When a number turns out wrong we revise it in the next report and name the assumption that failed, which is a different habit from defending it. Value per enquiry is also what makes an identical fix rank differently by category, since a fix worth a rounding error to a store is worth a quarter to a firm in legal marketing, and long-cycle B2B pipelines need a closed-won figure instead of a lead count for the arithmetic to be honest.

A ranked sequence of decisions, the evidence underneath each one, and an explicit list of what we recommend leaving alone. It is written for a reader who was not present while it was made, item by item, in the following order.

1. An executive summary for whoever gets through one page and never opens the rest. The three things limiting growth, the largest remaining opportunity, and what happens first.
2. The baseline. Every relevant number as it stood the day we started, recorded so the work can be retested against the original measurement instead of a moving one.
3. The ranked initiative list. Each entry carries five things. What to change, written concretely enough that somebody can start on it. Why, with the evidence attached. What it should return once done, with the assumptions sitting beside the number. What doing it takes, in effort and an approximate cost. And which kind of person has to carry it out.
4. The sequence, with dependency chains marked, so the order can be changed without accidentally breaking one.
5. The findings themselves, severity-rated, with the raw evidence attached. Annotated screens where the issue is visual, crawl and log extracts where it is technical, query exports for performance questions, and the assistant answers themselves where AI visibility is in scope.
6. The demand map. Terms grouped by what the searcher wants, each assigned to a page that exists or one that has to be built, with the accidental rankings flagged as the fastest gains.
7. The competitor read, including how each built what they hold and which of those methods is open to you.
8. The measurement plan. What to track, where it is recorded, what a good month looks like, and any tracking repair needed before the numbers mean anything.
9. The skip list, with a reason under each item. Usually the most quoted page in the document.
10. Confidence labels. Where we are certain, where a finding is a strong hypothesis with a named test attached, and where the evidence was thin enough to say so.

Two things about the way it is written are deliberate. It assumes you may execute it without us, so nothing in it needs a conversation we alone could hold and nothing needs tooling only we own. And where an item is actually a whole project and not a repair, the document says that out loud and prices it on its own. A content program, an authority program or a replatform each carry their own scope, and hiding one inside a line item would be a way of selling it quietly. You can see work that came out of these documents before deciding whether the format suits you.

Then it should work exactly as well, and the document is written on that assumption from the first page. Roughly a third of the roadmaps we deliver get executed by somebody other than us, which is a resourcing outcome and not a failure of the engagement.

You run all of it. Common where a capable in-house team was missing direction and not skill. We are out of it after the working session.

You split it. Content and campaign work stays internal, the specialist pieces come to us. This is the most frequent arrangement and usually the cheapest one that works.

Another agency runs it. Sometimes because they already hold the retainer and have context we do not. Sometimes because the honest answer is that they are better at the specific thing it calls for. Either way the handover is the same.

We run it. In which case the same people who did the diagnosis stay on the work, and the audit fee comes off the first phase.

For any of that to be true instead of a courtesy, five things have to hold.

• Findings written specifically enough to act on with none of us present
• No recommendation that leans on proprietary tooling
• Every technical item written as a specification a developer can implement, with the skill named, so you can tell whether it exists on your team
• Nothing withheld to manufacture a reason to hire us
• All of it yours, meaning the exports, the crawl files, the prompt sets, the value models and the accounts they came from

The fair objection is obvious, which is that a firm selling execution has a reason to find work for itself. The check sits in the parts of the document that send money elsewhere, meaning the skip list, the items assigned to your own team, and the occasions where the recommendation is a discipline we are weaker in. We have told clients to spend two quarters on conversion or on paid acquisition instead of search, and we have named a different firm for a specialism outside ours, both of which cost us revenue and are the only reason the rest of the advice is worth anything.

Handover is a session with whoever takes it on, walking through the reasoning behind it and not the conclusions alone, because a roadmap explained once executes better than one read cold. What we ask in return is a retest against the baseline at ninety days, whoever did the work. If you want execution under one roof afterward, the execution side sits with the same team, and if you would sooner just find out what is wrong first, tell us what you are dealing with.