TL;DR: To compare AI search optimization tools properly, ignore feature lists and test four things: whether the tool's numbers match your own manual checks, whether it stores full answer text and every cited URL, how many prompts you get per dollar, and whether it produces work you can act on. Most shortlists reorder once you run those four checks, and three of them cost nothing but an hour.
Key Takeaways
Verify accuracy yourself. Spot-check twenty prompts manually against the tool's data and record the match rate. Independent testing has found meaningful undercounting in parts of this category, so this is the check that matters most.
A mention flag is not citation tracking. You need the full answer text and every cited URL, including competitors', or you cannot diagnose why you lost.
Normalise to cost per tracked prompt. Headline prices are not comparable. At published entry tiers the real cost ranges from under a dollar to nearly four per prompt.
Decide if you need one tool or two. Visibility tracking and content production are separate categories. Very few products do both well, and buying one expecting the other is the common expensive mistake.
Local businesses need different criteria. If you serve an area, Google Business Profile and review signals drive AI recommendations more than your content does, and most comparisons skip this entirely.
Skip the AI metadata file features. Google's documentation states no new machine-readable or AI text files are needed, so a tool selling that as a headline capability is selling a checkbox.
Anchor targets assume your CMS generates slug IDs from heading text. Confirm on staging and adjust the hrefs if your platform slugifies differently.
What these tools are, and the three types
AI search optimization tools help your brand appear in answers from ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. The term covers three distinct product types that get compared against each other despite solving different problems, which is the first thing to sort out before you look at any feature table.
Related reading: Top Generative Engine Optimization Platforms
The three types
Visibility trackers | Content platforms | Technical and local tools | |
|---|---|---|---|
What they do | Measure whether engines cite you | Produce and publish content | Fix crawl access, schema, listings |
Core output | A rate and a work queue | Published pages | Shipped configuration changes |
Examples | Profound, Peec, Otterly, Scrunch | Blazly SEO, Surfer, Frase | Screaming Frog, Schema App, listings managers |
Prompt tracking | Yes | No | No |
Drafting at volume | No | Yes | No |
Entry price | About $29 to $500 | About $15 to $200 | Free to $250 |
A tracker tells you which prompts you lose. A content platform produces pages that might win them. A technical tool makes sure engines can reach those pages. Most serious programmes end up with one from the first two columns.
Which type you need
Run ten category prompts across three engines before shortlisting anything. Twenty minutes, no budget, and the result tells you which column to shop in.
What you find | The problem | The type to buy |
|---|---|---|
Not named at all, and logs show blocked crawlers | Access | Technical, often free |
Named but your URL never cited | Your pages have no quotable passage | Content, or just editing time |
Cited sometimes, competitors more | Coverage gaps | Content |
Cited well but you cannot see the trend | Measurement | Visibility tracker |
You cannot answer any of this | You have not measured | Neither yet. Measure first |
Two of five rows point at buying nothing immediately. That is the honest distribution, and it is why the comparison starts here rather than with a feature grid.
What this article covers
The criteria for comparing within a type, how to test accuracy yourself, and the dimensions most comparisons skip. For running a structured trial once you have a shortlist, see our guide to evaluating SEO tools, which covers the protocol in detail.
The seven criteria that change the decision
Most feature comparisons list twenty criteria and weight them equally. These seven are the ones that actually reorder a shortlist, ranked by how much they matter.
Ranked
# | Criterion | Why it ranks here | How to test it |
|---|---|---|---|
1 | Data accuracy | Every other number inherits any error here | Spot-check 20 prompts manually, record the match rate |
2 | Full answer text and cited URLs | Without it you cannot diagnose a loss | Ask to see one lost prompt end to end |
3 | Cost per tracked prompt | Headline prices are not comparable | Divide all-in fee by prompts at your real volume |
4 | Actionable output | A chart is not a work queue | Read ten recommendations and ask who could act on them |
5 | Engine coverage matching your buyers | Coverage counts are marketing | Check your own analytics for AI referrers first |
6 | Sampling frequency and variance | Answers vary run to run; one sample is noise | Ask how many runs per prompt per day |
7 | Export and portability | Your history is the switching cost | Ask for the export format in writing |
Why accuracy ranks first
Because it is invisible and it contaminates everything downstream. A tool reporting you in 11 of 20 prompts when manual checking finds 16 is not slightly wrong; every rate, trend and competitive comparison it produces inherits that gap.
No vendor will run this check for you, and it is an hour of work. The next section covers how.
Why engine coverage ranks lower than people expect
Tracking nine engines sounds better than four and usually is not, if your buyers use two. Check your own referral data and ask your sales team which assistants customers mention. Paying for coverage of engines nobody in your market uses is a common and quiet overspend.
The criterion nobody checks
Number seven. Your historical measurement data is the thing you cannot rebuild, and it is the reason switching later feels impossible. Ask at purchase time what export looks like, in what format, and how long you retain access after cancellation.
A vendor who will put that in writing is signalling something useful about how the relationship will go.
What is missing from this list
User interface quality, onboarding support and integrations. They matter for adoption and none of them changes whether the tool measures correctly. Weight them after the seven above rather than before.
Testing data accuracy yourself
This is the single most valuable hour in any tool evaluation, and no vendor will do it for you. The method is simple and the result is a number you can compare across candidates.
The method
Step 1. Pick twenty prompts from your tracked set, weighted toward category and comparative queries rather than branded ones.
Step 2. Run each manually in the engines the tool claims to cover. Record three things per prompt: whether your brand was named, whether your URL was cited, and which competitors appeared.
Step 3. Pull the tool's data for the same twenty prompts over the same period.
Step 4. Compare and record the match rate as a fraction. Nineteen of twenty is strong. Eleven of twenty is a problem.
What to record
Prompt | Manual: named? | Tool: named? | Manual: cited? | Tool: cited? | Match |
|---|---|---|---|---|---|
1 | Yes | Yes | Yes | Yes | Yes |
2 | Yes | No | No | No | No |
3 | No | No | No | No | Yes |
Run this table for all twenty and total the match column. Do it for every tool on your shortlist during their trials, which is why overlapping trials beats sequential ones.
Why discrepancies happen
None of these indicates bad faith, and all of them affect your numbers:
Different sampling times. Answers vary through the day and between runs.
Different prompt phrasing. Some tools normalise prompts before sending them.
Location and personalisation. Results differ by inferred location.
Detection method. Matching a brand name in text is harder than it sounds, particularly for generic or multi-word names.
Engine access method. Official API access and scraped consumer interfaces return different things.
That last point is worth asking about directly. A vendor using official APIs where available should say so readily.
What to do with the result
If two tools are close on price and one matches your manual checks substantially better, take the accurate one. If a tool undercounts badly, every downstream figure is wrong in the same direction, and you will make decisions on it for as long as you keep the subscription.
The shortcut version
No time for twenty? Do five, with one prompt you are confident you win and one you are confident you lose. A tool that gets either of those wrong has told you enough.
What good citation tracking looks like
Every tool in this category claims citation tracking. The term covers a wide range of capability, from a flag saying you were mentioned to the full answer with every source. Only the second lets you work out why you lost.
The capability ladder
Level | What you get | What you can do with it |
|---|---|---|
1 | A mention flag: yes or no | Know you have a problem |
2 | A visibility score | Watch a number move, with no formula |
3 | Mentions and citations reported separately | Tell awareness from source trust |
4 | Every cited URL, including competitors' | See who beat you and with what page |
5 | Full answer text stored per run | Read how you were described and why they won |
Level four is the minimum for a programme you intend to act on. Level five is what makes the weekly triage possible, because reading the answer tells you whether you lost on format, framing or absence.
Why mentions and citations must be separate
They are different events with different causes:
Pattern | What it means | What to fix |
|---|---|---|
Mentioned and cited | Working | Nothing |
Mentioned, not cited | Engines know you from third parties but find no quotable passage on your site | Page structure and openers |
Cited, not mentioned in the recommendation | You are a source but not an option | Positioning and constraint statements |
Neither | Not retrieved, or not in the consideration set | Crawler access, then entity data |
A tool that reports one blended figure cannot distinguish these, which means it cannot tell you which of four different fixes to apply.
The demo question that settles it
Send the vendor one prompt you know you lose, 48 hours before the call. Ask them to show you, live: the full answer text, every URL cited in it, and what their product recommends doing next.
A tool at level four or five does this in a minute. A tool at level one or two will show you a dashboard instead, which is itself the answer.
On sentiment
Sentiment classification is genuinely useful and frequently oversold. What matters is whether you can read the actual sentence describing your brand, not whether a model scored it positive. Ask to see the source text behind any sentiment score, and treat a score with no retrievable text as decoration.
The local and Maps dimension
If you serve a geographic area, most AI search tool comparisons are written for the wrong buyer. The criteria that matter for a local business barely overlap with those for a national SaaS, and almost no comparison says so.
Why local is different
When someone asks an assistant for the best provider in a city, the answer is assembled from business profiles, review platforms, directories and local best-of lists rather than from your blog. Your Google Business Profile feeds Gemini and AI Overviews directly.
SOCi's 2026 Local Visibility Index, analysing over 350,000 locations, found ChatGPT recommended roughly 1.2% of them against 35.9% appearing in Google's local pack, with about 45% overlap between the two groups. Winning Maps and being recommended by an assistant are related but far from the same thing.
What local buyers should compare on
Criterion | Why it matters locally | Found in |
|---|---|---|
City-level prompt tracking | Assistants rewrite near-me into a city name | Some trackers, by location setting |
Multi-location support | One profile per location, priced sanely | Listings managers, a few trackers |
Review monitoring and prompts | Review volume and recency drive recommendations | Reputation tools, rarely in AI trackers |
Directory and NAP consistency | Conflicting details make models hedge | Listings managers |
Profile completeness auditing | The highest-return free fix | Google Business Profile itself, free |
Notice how little of that column sits in a typical AI visibility tool. For a single-location business the honest recommendation is free tools plus effort rather than a subscription.
What to test if you are local
Ask each engine the five questions your customers actually ask, with your city named, twice each. Record whether you appear, how you are described, and which competitors and sources show up.
Then check whether the tool you are evaluating can reproduce that, with location set correctly. Many cannot, or treat location as an enterprise feature.
The multi-location trap
Per-domain or per-location pricing punishes you badly. Twenty locations at a per-location rate is not twenty times more valuable than one, and it will cost that. Look for tools pricing by prompt volume across locations instead, and price it against your real location count before any demo.
The honest recommendation for local
Complete your Google Business Profile, claim Bing Places and Apple Business Connect, make your details consistent everywhere, and build recent reviews. All free, all higher return than any tool, and most local businesses have not finished the first item. Buy software only once that is done and you still have a gap you can name.
Content and publishing integration
Some tools in this category add content generation and CMS publishing on top of visibility tracking. Whether you want that in the same product is a genuine decision rather than an obvious yes.
The case for one tool
The loop closes. The tool identifies a lost prompt, drafts a page targeting it, and publishes to your CMS without anyone exporting anything. For a small team that removes the handoffs where work dies, and it is one invoice and one login.
The case for two
Products are rarely strong at both. A tracker that added content generation usually has thin content capability, and a content platform that added tracking usually has thin tracking. Buying one suite often means accepting a weak half.
Two specialists also cost less than you would expect: a mid-market tracker plus a content platform frequently lands below a single enterprise tracker.
How to tell which half is strong
Ask which the company was founded to build. Then test the other half specifically during the trial, because that is where the thin capability will be.
If they started as | Test hardest | Likely weakness |
|---|---|---|
A tracker | The content output quality and CMS publishing | Generic drafts, limited publishing targets |
A content platform | The tracking accuracy and citation detail | Mention flags rather than full answers |
What publishing integration is worth
More than most people estimate. Moving a finished draft into a CMS by hand costs 15 to 20 minutes every time: reapplying headings, re-entering metadata, re-uploading images, fixing links. At forty articles a month that is over thirteen hours of pure administration.
If you publish at volume or manage several client sites on different platforms, direct publishing is frequently the single largest time saving on offer, larger than the drafting itself.
What to verify in a trial
Test publishing on a real article rather than a sample. Check that formatting survives, metadata arrives intact, internal links are preserved and images land correctly. Publishing integrations vary far more in quality than their feature lists suggest, and a broken one costs you the time it was meant to save.
The honest guidance
If your constraint is production, buy a content platform and keep tracking minimal or manual until you pass about fifty prompts. If your constraint is measurement, buy a tracker and accept that content remains your own work. Buying a suite to cover both is reasonable only once you have tested the half you care about less.
Technical features, real and otherwise
This is where feature lists get least reliable. Some technical capabilities in this category are genuinely valuable and some are checkboxes for things that do not work.
What is real
Feature | Why it matters | How to verify |
|---|---|---|
Server log analysis per AI user agent | The definitive answer on whether crawlers reach you | Ask to see status codes by agent, not just allowed or blocked |
Schema generation that validates | Entity identification genuinely helps | Generate markup and run it through the Rich Results Test |
Rendering checks | Content behind client-side JS may not be parsed | Ask whether it fetches with JS disabled |
Crawl diagnostics at scale | Finds blocks and indexing problems across templates | Run it on your own site during the trial |
What is not
Feature | The problem |
|---|---|
AI.json generation | Not a standard. No major provider documents support for it |
LLM-friendly robots.txt | Your robots.txt is a plain text file. There is no LLM-specific variant, only which user agents you name |
llms.txt generation | A real proposal with minimal adoption. Google's documentation states no such files are needed for its AI features |
Embedding optimisation | Embeddings are generated by the model from your text. No tool has access to tune them |
Google's documentation states plainly that site owners do not need to create new machine-readable files, AI text files or markup to appear in its AI features. A tool whose headline technical capability is generating those files is selling something with no documented effect.
The distinction that matters
Naming AI user agents correctly in your robots.txt is real and important. Generating a special file format for AI crawlers is not. Those two get conflated constantly, and only the first one changes anything.
What to ask a vendor
Ask which of their technical outputs are consumed by a documented standard and which are proposals.
A vendor who distinguishes schema.org from llms.txt accurately is being straight with you. One who presents both as equivalent requirements either has not checked or is counting on you not to.
The free alternative for most of this
PageSpeed Insights, the Rich Results Test and your own server logs cover rendering, schema validation and crawler access at no cost, and with Google's own checks against Google's own requirements. Pay for technical tooling when you need it at scale across many templates, not for the diagnosis itself.
The comparison table
This compares the three tool types against the seven criteria rather than naming products, because products change monthly and the structural differences do not. Use it to work out which column you are shopping in, then apply the criteria within that column.
Type against criteria
Criterion | Visibility trackers | Content platforms | Technical and local tools |
|---|---|---|---|
Data accuracy, testable | Critical. Spot-check always | Not applicable | Verifiable against your own logs |
Full answer text and cited URLs | The main differentiator | Not offered | Not applicable |
Cost per tracked prompt | The key price metric | Not applicable | Not applicable |
Actionable output | Varies widely. Most report only | The output is the action | Produces configuration changes |
Engine coverage | Compare against your buyers | Not applicable | Not applicable |
Sampling frequency | Ask runs per prompt per day | Not applicable | Crawl frequency instead |
Export and portability | Your history. Negotiate it | Your drafts, usually portable | Low lock-in |
Typical entry price | About $29 to $500 | About $15 to $200 | Free to $250 |
Who it suits | Anyone measuring seriously | Production-constrained teams | Everyone, mostly free |
How to use this
Read down your column. If you are shopping for a tracker, criteria one to three decide it and the rest are tiebreakers. If you are shopping for a content platform, the questions are entirely different: output quality, publishing reliability and editing time.
Named products, as of October 2026
Type | Commonly shortlisted |
|---|---|
Visibility trackers | Profound, Peec AI, Otterly.ai, Scrunch AI, Athena HQ, Semrush AI Visibility Toolkit, Ahrefs Brand Radar |
Content platforms | Blazly SEO, Surfer, Frase, Clearscope |
Technical and local | Screaming Frog, Schema App, WordLift, listings managers, plus free Google tools |
Pricing and positioning in the first row in particular has moved repeatedly this year, so verify against each vendor's own pricing page rather than relying on any comparison article, including this one.
The decision in one line
Pick the column from your diagnosis, then pick within the column on accuracy and cost per prompt for trackers, or on output quality and publishing reliability for content platforms. Everything else is a tiebreaker.
Pricing and what to ask
Headline prices in this category are not comparable, because allowances differ far more than fees do. Convert every quote to the same unit before you compare anything.
The conversion
Cost per tracked prompt =
all-in monthly fee / prompts you will actually track
(with every engine you need enabled)At published entry tiers this currently ranges from under a dollar to nearly four per prompt across the main trackers. The cheapest headline is frequently not the cheapest per prompt, and the most expensive headline sometimes is.
The questions to send in writing
Written answers are comparable and harder to revise later. Ask every shortlisted vendor:
What is the all-in monthly cost for the prompts and engines I need?
What is the prompt allowance, and the overage rate above it?
Which engines are included at that tier, and what does each extra cost?
How many sampling runs per prompt per day?
How many seats, and what does an extra seat cost?
Are there fees for API access, data export, or additional integrations?
Is there a setup or onboarding fee?
What is the renewal escalator, and will you cap it in writing?
Can I export full historical data, in what format, and how long after cancellation?
What is the term, notice period and auto-renewal mechanism?
The lines that surprise people
Line | Why it bites |
|---|---|
Prompt overages | The allowance is lower than you think once you segment by intent |
Per-engine pricing | The tier you need is rarely the tier advertised |
Per-domain or per-location | Punishing for agencies and multi-location businesses |
Annual-only billing | Removes your ability to leave quickly in a young category |
Export gated to higher tiers | Your switching cost, discovered at the worst moment |
The cost nobody quotes
Your own hours. Acting on what any of these tools surfaces runs 25 to 70 hours a month depending on volume, which at a loaded rate exceeds every subscription in the category.
That reframes the decision: a tool that genuinely reduces execution time is worth more than one that saves a few hundred a year on the invoice. Weight criterion four, actionable output, accordingly.
The rule
If two tools are within roughly $1,500 a year of each other, decide on accuracy and actionability. If one is substantially cheaper and matches your manual spot checks, take the saving.
A structured evaluation in eight steps
About three weeks alongside your normal work. The first three steps cost nothing and frequently remove candidates before you sit through a demo.
Step 1: Diagnose before shortlisting. Run ten category prompts across three engines and record whether you are named and which URLs are cited. This tells you which of the three tool types you are shopping for, and occasionally that you need none yet.
Step 2: Fix your prompt volume and engine list. Not what you would like to track, what you will. Both numbers drive every price comparison and most teams overestimate the first by double.
Step 3: Send the pricing questions in writing. Before any demo. Written answers are comparable across vendors, and anyone who will not answer in writing has told you something useful.
Step 4: Normalise every quote to cost per tracked prompt. Ten minutes, and it reorders most shortlists.
Step 5: Run the accuracy spot check on every finalist. Twenty prompts manually, compared against each tool's data, recorded as a match rate. This is the highest-value hour in the whole process.
Step 6: Bring one lost prompt to each demo. Ask to see the full answer text, every cited URL, and what the product recommends doing next. Fifteen minutes of real output beats an hour of slides.
Step 7: Test the integration that matters most to you. Publishing on a real article if that is your constraint. Log analysis against your own server data if that is.
Step 8: Negotiate export rights before signing. Full historical data, machine-readable, self-service, with a stated retention window after cancellation. Vendors rarely refuse because it costs them nothing, and it removes your biggest future lock-in.
The shortcut if you have a week
Steps 1, 4 and 5. Diagnosis, price normalisation and the accuracy check. Those three answer which type, which is affordable, and which is trustworthy, which is most of the decision.
What this does not cover
Running a full trial with a baseline, holdout and pre-set decision criteria. That is a separate protocol and it matters if you intend to measure whether the tool worked rather than just whether you liked it. Our guide to evaluating SEO tools covers it.
Document the decision
Write down why you chose, what you rejected and the renewal date with its notice period. In twelve months nobody will remember the reasoning, and that is how automatic renewals happen.
Six mistakes when comparing
1. Comparing across types. Putting a tracker and a content platform in the same table produces a grid of blanks and a decision based on feature count. Diagnose first, then compare within one type.
2. Trusting accuracy claims. Vendor data is presented confidently and is sometimes wrong. Twenty manual spot checks is an hour, and it is the only input in your comparison that comes from your own brand rather than from marketing.
3. Comparing headline prices. Allowances vary more than fees. Convert to cost per tracked prompt at your real volume or the comparison is meaningless.
4. Buying coverage you do not need. Nine engines sounds better than four and is worse value if your buyers use two. Check your own referral data and ask sales before paying for breadth.
5. Ignoring the local dimension. If you serve an area, most comparisons are written for a different buyer. Google Business Profile, reviews and directory consistency drive local AI recommendations more than any tool does.
6. Leaving export rights until you want to leave. Historical measurement data cannot be rebuilt, and it is the thing that makes switching feel impossible later. Negotiate it at signature, when it costs the vendor nothing to grant.
The pattern
Four of six come from comparing the wrong things rather than comparing badly. The diagnosis in step one prevents three of them outright, and it costs twenty minutes.
The one that costs most
Mistake two, because it is invisible and permanent. A tool that undercounts will produce a confident, consistent and wrong picture for as long as you keep it, and every decision you make on that data inherits the error. An hour of manual checking during a trial is the cheapest insurance available in this entire category.
FAQs
How do I compare AI search optimization tools?
Diagnose which of three tool types you need, then compare within that type on four things: accuracy against your own manual spot checks, whether you get full answer text and cited URLs, cost per tracked prompt at your real volume, and whether the output is actionable.
How do AI search tools differ from traditional SEO software?
Traditional tools track keyword positions on results pages. AI search tools sample prompts against engines like ChatGPT and Gemini, record whether your brand is named and which URLs are cited, and report that as a rate rather than a position. The technical foundation underneath is shared.
How do I check whether a tool's data is accurate?
Run twenty prompts manually across the engines it covers, record whether you were named and cited, then compare against the tool's figures for the same period. Record the match rate. Nineteen of twenty is strong, eleven of twenty is a problem.
What is the difference between mention tracking and citation tracking?
A mention is your brand name in the answer text. A citation is your URL in the source list. They are separate events with different causes, and a tool reporting one blended figure cannot tell you which of four different fixes applies.
How much should these tools cost?
Visibility trackers run roughly $29 to $500 a month at entry, content platforms about $15 to $200, and most technical diagnosis is free. Compare on cost per tracked prompt rather than headline price, since allowances vary far more than fees do.
Do these tools help with Google Maps and local visibility?
Some offer location settings, but the criteria that matter locally are different. Google Business Profile completeness, review recency and directory consistency drive AI recommendations for local queries more than any tracking tool, and most of that work is free.
Do I need separate tools for tracking and content?
Often yes. Few products are strong at both, and two specialists frequently cost less than one enterprise suite. Ask which half the company was founded to build, then test the other half hardest during the trial.
Are AI.json and llms.txt files worth paying for?
No. Google's documentation states no new machine-readable files, AI text files or markup are needed to appear in its AI features. Naming AI user agents correctly in your robots.txt is real and important; generating special file formats is a checkbox.
Making the comparison
The work that decides this is mostly free and mostly yours. Diagnose which tool type you need with ten prompts, normalise every quote to cost per tracked prompt, and spot-check accuracy against your own manual checks. Three hours total, and they matter more than any feature grid.
Then compare within a single type. Trackers compete on accuracy, citation detail and cost per prompt. Content platforms compete on output quality, publishing reliability and editing time. Those are different conversations and running them together produces a decision nobody can defend later.
When to buy nothing yet
You have not run the diagnosis. Twenty minutes tells you whether you have a problem and which kind.
Your crawler access is blocked. Fix that first. It is usually one line and it dominates any measurement you would buy.
You are a single-location local business. Complete your Google Business Profile, claim Bing Places and Apple Business Connect, build reviews. Free, and higher return than any subscription.
Under fifty prompts and no production backlog. Manual checking and your existing process are sufficient for another quarter.
If production is the constraint
If your diagnosis shows you know which prompts you lose and cannot get the pages written, that is a content constraint rather than a measurement one. Blazly SEO covers research, cluster planning, drafting, on-page scoring and direct publishing to HubSpot, WordPress and Webflow from about $39 a month.
It does not track prompts or attribute citations, so it sits alongside a tracker rather than replacing one. Trial it on a real cluster and time the whole process including your editing.
Summary
To compare AI search optimization tools, start by working out which of three types you need, because visibility trackers, content platforms and technical tools solve different problems and get compared against each other constantly. Within a type, rank on data accuracy tested against your own manual checks, full answer text with every cited URL, cost per tracked prompt at your real volume, and whether the output is something a person can act on. Check the local dimension if you serve an area, since the criteria there are different and mostly free. And discount any tool whose headline technical feature is generating AI metadata files, because Google's documentation says you do not need them.