A calibration curve on raw response quietly assumes that the amount of analyte reaching the detector is the amount you meant to deliver. On a good day it nearly is. But you seldom draw exactly 1.00 µL into a syringe twice running; a little sample clings to a pipette or is lost in an evaporation step; a plasma's transport efficiency drifts as the day warms and the cones foul. Each of these multiplies your signal by a factor you did not choose and cannot see — and an external curve reads that wobble straight into the answer. The internal standard is the trick that makes such factors cancel: put a known companion compound into every sample and every standard, and quantify not by the analyte's raw signal but by its signal relative to the companion's. To the extent something scales the analyte and the companion by the same factor, a ratio does not see it.
The idea in one line
Add the same known amount of a reference compound — the internal standard — to every calibration standard, sample, blank and QC, by the same technique. Then calibrate the ratio of analyte response to internal-standard response against concentration, and read unknowns from that ratio. Because the internal standard sits in the vial with the analyte, riding the same injection, plasma, and detector on the same run, a factor that scales one tends to scale the other — and the ratio drops it.
Here f is a shared, run-to-run fluctuation — injected volume, transport efficiency, a drifting gain. What survives the ratio is k_a / k_IS, the analyte's sensitivity relative to the internal standard's: the relative response factor (RRF). It is not a number you lock once and forget: relative response drifts too, so you build the ratio calibration from standards on each sequence and confirm it still holds — a continuing-calibration check — before you trust it.
Two things have to hold, and they are different in kind. For f to cancel at all, it must be a single multiplicative factor acting equally on both channels. Separately — and this is an accuracy assumption, not a cancellation one — the RRF you measured from standards is only right for samples if k_a/k_IS is the same in both, i.e. the matrix does not tilt the analyte and internal standard relative to each other. On the first: f is genuinely common for things that act on the whole vial at once — injected volume, bulk transport from that vial. It is not automatically common for effects that hit the two compounds differently: inlet/split discrimination and liner activity in GC, a co-eluting matrix suppressing one but not the other in LC-MS, mass-dependent transmission in ICP-MS, a detector gain that varies with mass or compound. Every such difference is variation the internal standard cannot see — which is exactly why the requirement below, "as analyte-like as possible," is not style advice but the condition the equation runs on.
What the internal standard has to be
Absent from the sample. If the compound you add is already present, your "known amount" is wrong and so is the ratio. Choose something the sample cannot contain.
Cleanly distinguished in the signal — in one of two ways. If the internal standard is a different compound, it must be well resolved from the analyte and from interferences, a clean distinct peak (and a distinct m/z on a mass spec). If it is an isotope-labelled version of the analyte, the opposite is wanted: it co-elutes on purpose and is separated only on the mass axis, with the pair chosen so neither channel bleeds isotope signal into the other.
As analyte-like as you can manage. The internal standard cancels a factor only to the degree it feels that factor the same way the analyte does — the point the equation made above. A companion that evaporates, extracts, and ionizes like the analyte tracks its losses faithfully; one that behaves differently tracks them poorly, cancels less, and can add scatter of its own. The closer the chemistry, the better — which points at the limiting case below.
Added at the right stage, and the right level. Add it before the steps whose variation you want to cancel — before the run to catch injection and instrument drift, before the extraction or dilution to catch preparation loss. Dose it to a response of similar magnitude to the analyte's; comparable size keeps both in the linear range with good signal-to-noise (a working rule of thumb, not a precision theorem).
An internal standard cancels a shared, multiplicative fluctuation that scales analyte and internal standard by the same factor: injected volume, bulk transport, uniform loss, a drifting gain. It does not cancel anything that touches the analyte but not the companion — a matrix that suppresses the analyte's ionization but not the internal standard's, a selective loss, an interference on the analyte's channel. Nor does it remove an additive background (a blank's job), and it does not rescue a nonlinear curve. A badly matched internal standard, tracking the wrong thing, adds noise instead of removing it — which is why choosing it is a real decision, not a formality.
Surrogate or internal standard?
The names are really about function. An internal standard is the compound whose ratio goes into the concentration formula — it corrects the result. A surrogate (recovery standard) is there to watch recovery through preparation; its recovery is reported as a QC flag but usually does not go into the number. Timing tends to follow that split — surrogates in before prep, a pure-instrument internal standard just before analysis — but the function is the definition, not the clock. And the two blur at the top: in isotope dilution and most regulated bioanalytical work (ICH M10), the internal standard is a labelled analog added at the start of prep, correcting both the preparation and the instrument at once — an internal standard by function, added like a surrogate. (The EPA-organics vocabulary is one common convention; term usage varies elsewhere.)
The near-perfect internal standard — with caveats
Follow "as analyte-like as possible" to its limit and the companion becomes a version of the analyte itself, differing only in isotope — an isotopically labelled analog. Chosen well it co-elutes and tracks nearly every loss and matrix effect the analyte suffers, and using it as the internal standard for a mass spectrometer is the basis of isotope dilution — the most accurate everyday MS approach and the norm in bioanalytical and clinical work.5 Two caveats keep it from being magic. First, "differs only in mass" is an idealisation: deuterated analogs can shift retention slightly (a chromatographic isotope effect) and occasionally extract or ionize a touch differently; ¹³C and ¹⁵N labels sit closer to identical. Second, the classical, metrological isotope-dilution MS that ranks near a primary method — a gravimetric spike of a certified labelled standard, quantified from a measured isotope ratio, in specific methods — is not the same thing as running an ordinary calibration curve with a labelled internal standard, though both are called "isotope dilution." Its real failure modes: unlabelled impurity in the spike, too small a mass shift so the analyte's natural isotopes overlap the label, spectral overlap (chlorine and bromine patterns are notorious), cross-talk between channels, and incomplete equilibration of spike with sample. It is the internal-standard idea taken as far as it goes — powerful, not automatic.
An analyte is run against a fixed internal standard on GC-FID; the manual injection volume varies shot to shot. Three replicate injections of one standard (areas, arbitrary units):
| Injection | Analyte area | IS area | Ratio A/IS |
|---|---|---|---|
| 1 | 4820 | 9910 | 0.4864 |
| 2 | 5110 | 10480 | 0.4876 |
| 3 | 4590 | 9450 | 0.4857 |
The raw analyte area has a relative standard deviation near 5% across the three shots; the internal-standard area varies right along with it, run for run, because both rode the same injection — so the ratio's RSD is about 0.2%. This is not a calibration (there is no RRF, no unknown here) — it is the mechanism made visible: the part you could not control divides out. One habit it implies — watch the internal-standard signal itself, against an acceptance window. A few percent of correlated jitter, as here, is the method working; but an internal-standard area that falls outside that window is a flag — the ratio may still "correct," yet it signals that f is no longer simple shared volume or gain (or that linearity or identity has broken).
Where it sits in the family
Three pieces, three tools. An external standard is the default when the instrument is stable and the matrix is clean. Standard addition is for a matrix effect you cannot reproduce in a standard. The internal standard is for run-to-run variability — injection, prep, drift, transport — that would otherwise ride into the answer. In plasma spectrometry an internal standard is the routine control for the physical run-to-run effects (drift, nebulisation and transport, sample viscosity); it is not a cure-all there, since mass-dependent and easily-ionized-element effects need an internal standard matched in mass and ionization potential — often several across the mass range — and a genuinely hostile matrix still calls for matrix matching or standard addition. These tools are not rivals: an internal standard almost always rides along with an ordinary calibration, and can be layered onto standard addition too (plot the analyte/IS ratio versus added amount). The external curve gives you the map; the internal standard keeps it honest from one run to the next.
The judgement
An internal standard is a deliberate bet that a chosen companion will suffer what your analyte suffers, and it pays off only to the degree that bet is sound. Choose a compound genuinely like the analyte, resolved from it (or a co-eluting labelled twin), absent from the sample, added at the right stage and level, and it removes run-to-run variation that careful technique alone cannot. Choose a convenient-but-dissimilar one and you have added a second source of scatter and called it a control. Knowing the difference — and knowing which variability the internal standard can and cannot see — is the skill; the ratio only rewards a companion well chosen.
A foundations piece — the reasoning behind internal standards, not a validation protocol. For internal-standard selection, calibration and continuing-check criteria, and recovery limits, work from the relevant EPA/ASTM method, ICH M10 for bioanalytical work, and your laboratory's quality system.