Ahrefs vs Semrush for SEO and AEO? Choose the Job First

Two tabs, two checkout pages, both around a hundred and forty dollars a month before you add anything, and feature lists that read almost identically. Keyword research, backlinks, rank tracking, site audit, and now an AI visibility module on both. Three days in and you are no closer, because the comparison you are running cannot be won.

So here is the answer. Ahrefs and Semrush overlap so heavily on the four classic SEO jobs that for most buyers the real choice is not which tool is better, but which dataset you will open every Tuesday and what the AEO module is actually measuring. Pick the job first and the tool falls out of it. I am not going to name a winner; what follows separates the numbers in each product that are measurements from the ones that are models or samples, and shows where the pricing costs more than the sticker.

Five jobs, and the two you will actually do weekly

An SEO platform is five products sold as one: keyword and topic research, backlink data, rank tracking, site auditing, and now AI visibility tracking. What nobody says at the point of sale is that almost every buyer uses one or two heavily and pays for all five.

An in-house marketer at a services firm lives in keyword research and rank tracking and opens the backlink tool twice a year. A link builder ignores the site audit entirely. A technical SEO on 40,000 URLs lives in the crawler. So the question is which of the five is your weekly job, and whose version of it you prefer. Check the rest of your stack too: if you already pay for a dedicated crawler, the site audit is a duplicate you are funding.

Volume and difficulty are models, not measurements

Third-party search volume and keyword difficulty are estimates produced by a vendor’s model, not counts of searches. The two products disagree on the same keyword because they run different models over different proxy data. That is not a defect. It is what the numbers are.

Google publishes per-keyword search counts to nobody. What exists publicly is Keyword Planner, built for advertisers; as Ahrefs’ own glossary notes, “In Google’s Keyword Planner and many SEO tools, Search Volume is the annual average value.” Vendors refine that with clickstream panels and modelling: “More advanced tools like Ahrefs use different methods to refine the data, which results in a difference in reported search volume.” Semrush’s data article says keywords come from Google SERPs “for hundreds of millions of the most popular keywords”, that updates “can range from once a day to once a month”, and that traffic estimates use “clickstream data through our Neural Network algorithm”.

Neither vendor publishes a complete volume methodology. I went looking and found no panel size, no geographic weighting, no error bars, so nobody outside the companies can audit those figures. Do not imply otherwise in a client report.

Difficulty is more transparent and more limited. Ahrefs states the method: “We pull the top 10 ranking pages for your keyword and count how many websites link to each… Then we plot the result on a logarithmic scale from 0 to 100”, with the caveat in writing: “Keyword difficulty is always only an estimation because Google does not disclose all its ranking factors.” A KD of 42 describes referring domains to today’s top ten. It says nothing about intent match, page quality or brand.

So use vendor volume as a relative signal. “A looks roughly ten times B” is usable; “1,900 searches a month” as a forecast input is not. And for anything you already rank for, use Search Console, which measures your own property instead of modelling the market: clicks are “the number of times a user clicked your site from Google Search results”, impressions “how many times your site appeared in Search results”, average position “the average position of the topmost result from your site”.

Google’s stated limits on that data matter as much. Recent rows are provisional, since “the newest data can be preliminary, meaning it’s still being collected and might change in the next few hours”. Chart and table totals can disagree over aggregation, “property vs. page”. And it holds only queries you already appear for, with Google warning that “even if a query appears in your list, you might not see your site in results if you run the same query in Google Search”. A partial view, and still the best you have.

Backlink indexes: where Ahrefs and Semrush genuinely differ

This is the one place the two are not substitutes, for a mechanical reason. Each vendor runs its own crawler, so each holds a different set of links, found at different times and retained under different rules. Nobody crawls “the web”; each crawls as much as its infrastructure reaches, in its own order.

Both publish self-reported figures. Ahrefs’ data page, when I looked on 12 September 2026, showed “5 M Pages crawled every minute”, “493.9 B Pages in index”, “35.0 T External backlinks history” and “209.5 M Domains (post-vetting)”, describing an index “updated with fresh data every 15–30 minutes”. Semrush’s documentation states: “Our backlink data comes from our own proprietary backlink crawler. This crawler combs through approximately 10 billion web pages daily, adding new links to our database of over 43 trillion backlinks.”

Resist the arithmetic. Those totals are not the same unit: one is external backlinks history, the other a database of backlinks, and neither publishes an audited definition of what enters or what is kept after a link dies. Comparing 35 trillion with 43 trillion tells you which marketing page carries the bigger number. The test that tells you something takes twenty minutes.

Write down ten to fifteen links you know exist: a supplier’s dealer page, a trade body directory entry, a local news mention, a podcast page, a client case study, a recent PR hit.
Include three or four from the last fortnight, because freshness matters more than size for link building.
Look each up in both trials: present or absent, and if present, when first seen.
Repeat for your two closest competitors, whose profiles are what you will actually mine.

Eleven hits against six is your answer, and it is yours rather than a reviewer’s. Coverage varies by language, region and how obscure the linking sites are, which is why a generic comparison never settles this.

A tracked position is a sample, not a fact

The number in your rank tracker is one observation of a query with no single correct answer. Google is explicit that location shapes results: “your current location is used to give you more helpful results”, estimated from device location, addresses on your account, prior activity and your connection’s IP address. History plays in too, in Google’s own example, “if you search for coffee shops in Chelsea and then nail salon, Google might show nail salons in Chelsea”. Add device differences, SERP features pushing organic down the page, and constant live testing. Ranking is a distribution, and your tracker draws one sample from it.

That makes rank tracking good for direction and useless as fact. Configure it so the sampling is at least consistent:

One location per keyword group, set once, as tight as the tool allows and matched to where your customers are. Not the country default if you sell in one city. Same language and search engine for the life of the project.
Separate mobile and desktop. Do not average them, and do not switch later; the series breaks.
A frozen keyword list, because adding or removing keywords changes the average visibility score and you will read your own edit as a result.
Three-week trends, not daily wobble. Both sell daily updates; Semrush even expresses plan limits as keywords “to track daily”. Daily data is not daily meaning.

Keep a dated note of every configuration change. Six months on, that note is the difference between explaining a step change and guessing at one.

What the AEO modules actually do today

The mechanism is plainer than the marketing. An AI visibility module runs a set of prompts against a set of assistants on a schedule and records whether your brand was mentioned in the answer text or cited with a link. It is a sampling exercise, not a traffic measurement.

The two sample very differently, and that is verifiable from their own documentation. Ahrefs’ Brand Radar covers “AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, Copilot”, with Claude available for custom prompts, and builds prompts from search data rather than by hand: real queries from its keyword database, expanded via People Also Ask and semantic fanout, with the resulting 459M+ prompts run through each platform. Question sets are “re-tested monthly on a 90-day reporting window”. It reports mentions (“how often your brand name appears in AI answer text”), citations (“how often AI links to your pages or domains”), AI Share of Voice, and estimated impressions weighted by search volume.

Semrush’s documentation describes tracking against “LLMs like ChatGPT and Google’s AI Mode”, states “Prompt tracking: updated daily”, and lists visibility overview, competitor research and prompt research as “updated daily on a rolling basis”. It reports an AI visibility score built on brand mentions against a competitor median, plus share of voice, sentiment and citations.

Monthly re-testing across a 90-day window and daily tracking of prompts you chose yourself answer different questions: broad and slow against narrow and fast. Neither cadence is the correct one, which is why the choice should follow from whether you need coverage or responsiveness. Three things to hold on to, none of which appeared in the documentation I read.

The answers are non-deterministic. The same prompt run twice can return different sources, different wording and different competitors, because responses are sampled rather than looked up, retrieval is live, and vendors ship changes constantly. A score built from one run per prompt carries sampling error nobody reports to you.

The prompt sets are not comparable. Hundreds of millions of machine-expanded questions and twenty-five prompts you wrote yourself are different instruments. Numbers from one cannot sit beside numbers from the other.

No vendor can see the assistants’ own traffic. Not Ahrefs, not Semrush, not any AI visibility startup. Mentions and citations are observations of answers, not of visits, and visits have to be measured on your own property, which has its own structural problems; that is written up separately in how to measure AI traffic to your site. Buy the module to watch share of voice in answer space, not as a traffic report.

Where the money actually is in the pricing

What follows is the pricing model shown on each vendor’s own pricing page in US dollars when I checked on 12 September 2026. Both change prices and plan contents, and what you are shown varies by region and currency, so verify your own numbers at checkout.

Ahrefs lists Lite $129/month, Standard $249, Advanced $449 and Enterprise $1,499 with an annual commitment required. Each of the first three includes one user; extra users are $40, $60 and $80 a month, capped at two, five and ten. Projects run 5, 20, 50; tracked keywords 750, 2,000, 5,000; crawl credits 100,000, 500,000, 1,500,000 a month. Historical depth is tiered at six months, two years and five years. Pay-as-you-go credits cover crawls, exports and API units past plan limits, and Brand Radar AI is priced separately from $199/month with prompt packages listed from $50 to $250/month. Annual billing is advertised as “Pay annually, save up to 17%”.

Semrush lists SEO at $139/month or $117.33 billed annually, Starter $199 or $165.17, Pro+ $299 or $248.17, Advanced $549 or $455.67, also advertised as “save up to 17%”. Limits appear as “websites to monitor” and “keywords to track daily”: 5 and 500 on SEO and Starter, 15 and 1,500 on Pro+, 40 and 5,000 on Advanced. AI work is metered in prompts, at “50 prompts to track daily” on Starter, 100 on Pro+ and 200 on Advanced, with Starter also listing “1 domain for AI brand performance” and “300 AI visibility reports per day”. Extra users start at $45/month. Semrush’s knowledge base article on the AI visibility toolkit describes a base allowance of 25 prompts, extra prompts at 50 for $60/month, and additional brand performance domains at $99 each.

Note that mismatch: the pricing page and the knowledge base state different base prompt allowances. Not scandalous, just what happens when a product ships fast, and a good reason to get the allowance confirmed in writing rather than inferred from a plan card. The traps, in the order they catch people:

Seats. Both include one user and charge per extra one, so a three-person team pays well above the headline every month. Price your real headcount, not one shared login: sharing is usually a terms violation.
Projects and tracked keywords. Agencies hit those caps long before any data limit. Count live clients and their keyword lists before picking a tier.
Credits and units. Crawl credits, export limits, API units and report allowances are all rationed. The month you run a big site audit is the month you find the ceiling.
History as an upsell. Six months of data on an entry plan is a real constraint if you need year-over-year comparisons.
The annual commitment. This is where the real cost sits. Roughly 17% off looks obvious and locks twelve months to a product you have used for two weeks. Pay monthly for the first quarter; the premium is about a fifth of a year’s fee for the right to be wrong.

The two-week trial that settles it

Feature grids cannot decide this, because both grids are nearly full. Task timing can. Pick three tasks you will genuinely do weekly, do each in both products, and score two things only: how long it took, and whether you trusted the output.

“Find fifteen realistic new target queries for this service page.” End with a list you would brief a writer from.
“Who links to my three closest competitors, and which of those links could I plausibly get?” End with named prospects, not a CSV.
“Explain why this page’s traffic dropped last month.” Cross-reference against Search Console. This exposes more product quality than the other two combined.
“Produce the client-ready one-page report.” If you bill anyone, reporting speed is a weekly cost.
“Crawl the site and give me the ten issues worth fixing.” Judge by how much output is noise you have to filter.

Time each run and write one sentence on whether you believed it, then run the backlink spot-check from earlier. You end up with a table of real minutes, and the decision usually makes itself. If it does not, the two are interchangeable for your job: buy the cheaper monthly plan and stop spending your time on this. One rule during the trial. Do not play with features you will not use.

When you should buy neither

Some of the people asking this should not spend the $139 yet, and nobody selling either product will say so.

Free coverage goes further than most people think. Search Console gives you real clicks, impressions, average position and the indexing status of every page, measured rather than modelled. Bing Webmaster Tools adds independent crawl and query data. Google Business Profile is the whole game for a local business and costs nothing. Google Trends handles relative demand and seasonality. Between them you can pick targets, find opportunity pages and diagnose indexing problems with no subscription, in the order set out in the piece on starting SEO with no budget.

My rule of thumb, offered as judgement rather than a finding: if fewer than about thirty of your pages earn impressions in Search Console, you will get more from fixing pages than from measuring them. At that size the constraint is almost never information. It is that four services share one page, the pricing page does not exist, the titles do not match what people type, and half the site is not indexed. A subscription shows you that slowly. A morning of editing fixes it.

The other case for neither: you need one dataset badly and nothing else. Purely link prospecting? A backlink-only subscription can cost less. Tracking forty keywords? A standalone tracker is cheaper. Platforms pay off when you use three or more of the five jobs. Buy one the moment you can finish this sentence: “on Tuesday morning I will use it to…”. Otherwise it becomes a bookmark you feel guilty about.

If you would rather have this work done than researched, that is what we do. AB7 Solutions runs SEO, AEO and GEO programmes end to end: the keyword and content strategy a platform is only supposed to inform, a fixed-prompt AI visibility audit built so the numbers survive the non-determinism problem above, and the website, application and API work when the site itself is what holds results back. Hire us for the programme and the tool choice becomes our problem rather than a decision you have to defend. If your Search Console says you have twenty pages and should spend the money fixing them instead, we will say so and hand over the list. Call +1 321 341 7733, email ab@ab7solutions.com or director@ab7solutions.com, or start at www.ab7solutions.com.

Sources, all checked 12 September 2026: Ahrefs pricing, Ahrefs data and index figures, Ahrefs on Keyword Difficulty, Ahrefs on search volume, Ahrefs Brand Radar; Semrush pricing, Semrush data sources, Semrush AI Visibility Toolkit documentation; Google Search Console Help, Performance report; Google Search Help, Google Search location settings. Index sizes and crawl rates are vendor self-reported and unaudited, and no public methodology for either vendor’s search volume estimates was available on the pages cited.